Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Electron Transport in Semiconductor Nanowires and Electrostatic Force Microscopy on Carbon Nanotubes Ph.D. Thesis Thomas Sand Jespersen February 2007 Niels Bohr Institute Faculty of Science University of Copenhagen Denmark Thesis Advisor: Assoc. Prof. Jesper Nygård Electron Transport in Semiconductor Nanowires and Studies of Carbon Nanotubes using Electrostatic Force Microscopy Ph.D. Thesis c °Thomas Sand Jespersen 2007 e-mail: [email protected] Niels Bohr Institute Nano-Science Center Faculty of Science University of Copenhagen Universitetsparken 5 DK-2100 Copenhagen O Denmark ii Abstract A study is presented of the electronic transport properties of semiconducting Indium Arsenide nanowires. The wires are grown by molecular beam epitaxy and have diameters ∼ 50 − 70 nm and lengths ∼ 3 − 5 µm. A scheme is developed for making electrical devices in a field effect transistor geometry where metal contacts to the wires are lithographically defined. The transport characteristics are measured as a function of bias voltage, temperature, gate potential, and magnetic field. Measurements have been performed on devices in different transport regimes enabled by differences in the transparency of the nanowire-electrode interfaces. Devices with large barriers exhibit clear Coulomb blockade and the level structure is analyzed, showing evidence of transport through spin-degenerate levels. The quantum dot structure is highly regular and the sequential addition of ∼ 150 electrons can be followed. The dependence on electrode separation is investigated showing that the confinement of the quantum dot is determined by the lithographically defined contacts. For larger transparencies, higher order transport processes are found to contribute significantly, and the phenomena of cotunneling and the Kondo effect are observed. In devices with open contacts the nanowires act as phase-coherent diffusive conductors and electron interference gives rise to conductance fluctuations and weak (anti) localization in the magnetoconductance. A section is devoted to measurements on nanowire devices contacted with leads of a Ti/Al/Ti trilayer turning superconducting below the transition temperature Tc ∼ 750 mK. Results from the three transport regimes described above are presented, with emphasis on the intermediate coupling regime where the Kondo effect is observed for temperatures above Tc . Due to the competition of the Kondo effect and superconductivity the zero-bias Kondo peaks are (in most cases) suppressed once the device is cooled below Tc . However, the Kondo effect is found to have a pronounced effect on the sub-gap structure, greatly enhancing the conductance peak at a finite bias voltage of ∆/e (where ∆ is the superconducting energy gap). By comparing with results obtained on carbon nanotubes the effect is shown to be a general quantum-dot phenomena and a possible explanation is discussed. The problem is currently under theoretical consideration. Also presented are investigations of single wall carbon nanotubes using a scanning probe technique known as electrostatic force microscopy (EFM). The superiority of the technique for fast, large scale characterization of nanotube distributions is demonstrated and it is shown to convey also information about defects on a single nanotube level, as well as the interplay of nanotubes with static charges on the substrate. Finally it is shown that EFM can be used for mapping the three-dimensional orientation and position of nanotubes embedded in a polymer matrix. iii iv Preface This thesis is submitted to the Faculty of Science at the University of Copenhagen in partial fulfillment of the requirements for the Ph.D. degree in Physics. The experimental work on carbon nanotubes and semiconducting nanowires presented in the thesis has primarily been carried out at the Ørsted Laboratory, Niels Bohr Institute. The aim of this three-year project was to initiate the studies of semiconductor nanowires in the nano-physics group utilizing the existing experimental facilities for semiconductor processing and lowtemperature measurement setups for investigations of transport properties. I am sincerely grateful to my supervisor Jesper Nygård for introducing me to the world of mesoscopic transport and for invaluable guidance in all aspects of my project, from overall strategies and interpretation of results to the details of how-to-run-the-cryostat. Also, I wish to thank Prof. Poul-Erik Lindelof for many inspiring discussions and physical insights. I am very grateful to Claus Sørensen and Martin Aagesen. Without your pioneering work on growing nanowires by molecular beam epitaxy this project would have developed much differently. For inspiring discussions (on physics-related as well as non-physics related issues), for always being willing to help, and for our everyday lunch-breaks I am very grateful to my co-workers Kasper Grove Rasmussen, Jonas Rahlf Hauptmann, Anders Mathias Lunde, Henrik Ingerslev Jørgensen, Jeppe Holm, Martin Aagesen, Søren Stobbe, Pawel Utko, Ane Jensen, Magdalena Utko, and Brian Skov Sørensen - I hope we will be able to keep in touch and get together once in a while even though we will probably soon find ourselves scattered around the globe. Thanks to Claus Sørensen, Nader Payami and Inger Jensen for all your technical support and for keeping the lab running smoothly, and to Carsten Hyldebrand Mortensen, at the workshop, for all your skilled assistance in designing and constructing the CVD lab. On the theory side, I wish to thank Jens Paaske, Karsten Flensberg and Brian Møller Andersen for taking an interest in our experiments, and for working on a theoretical description of the non-equilibrium properties of a Kondo-dot contacted by superconductors. For granting me access to their Raman spectroscopy laboratories, I which to thank Prof. Eleanor Campbell (Gothenburg University, Sweden) and Prof. Ole Faurskov Nielsen (Institute of Chemistry, University of Copenhagen). During the course of the project I stayed six very stimulating months in the group of Prof. Charlie Marcus, Harvard University, and I am very grateful to Prof. Marcus for granting me this opportunity and for his many insightful and encouraging thoughts on the nanowire project. Also, I am grateful to Hans-Andreas Engel and Emmanuel Rashba for taking an interest in our v experiments. I wish to thank all the people in the Marcus Lab for making my stay in Cambridge a very memorable one and I especially wish to thank Jimmy Williams for many enjoyable hours in the lab, and for giving me an inside view on the American culture. Thomas Sand Jespersen, Copenhagen, February 2007 vi Contents 1 Introductory comments 3 2 Electron Transport in Semiconducting Nanowires 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 2.2 InAs nanowires . . . . . . . . . . . . . . . . . . . . . 2.3 Device preparation and experimental details . . . . . 2.4 Nanowire devices in the closed dot regime . . . . . . 2.5 Kondo physics in InAs nanowires . . . . . . . . . . . 2.6 Conductance Fluctuations in InAs nanowires . . . . 2.7 Nanowire devices with superconducting leads . . . . 2.8 Conclusion of nanowire studies . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 5 13 18 23 39 50 55 69 72 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Electrostatic Force Microscopy on Carbon Nanotubes 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Experimental Details . . . . . . . . . . . . . . . . . . . . 3.3 The EFM technique . . . . . . . . . . . . . . . . . . . . 3.4 Identifying defects using EFM . . . . . . . . . . . . . . . 3.5 Studying surface charges using EFM . . . . . . . . . . . 3.6 EFM for assessment of embedded CNT’s . . . . . . . . . 3.7 Conclusion of EFM studies . . . . . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 . 79 . 84 . 87 . 94 . 99 . 105 . 112 . 114 A InAs nanowire growth NBI73 119 B Details of nanowire device fabrication. 121 C Type-1 devices in the closed-dot regime 124 D Additional Kondo data 125 E EFM parameters 127 F Publication list 129 1 2 Chapter 1 Introductory comments The work presented in this thesis has been carried our during the last three years. It can be categorized into two rather separate areas and a chapter has been devoted to each; Chapter 2: ”Electron transport in semiconducting nanowires”, and Chapter 3: ”Electrostatic Force Microscopy on single walled carbon nanotubes”. Each chapter is self-contained and can be read separately, however, a few references are made from chapter 3 to the material in chapter 2 and thus the chapters are preferentially read in the order in which they are presented. The studies of electron transport in InAs nanowires, presented in chapter 2, constitute the subject of the originally proposed project and contains the main body of material. The two chapters contain their own introductory sections but before continuing, a comment should be made on the link between the presented material: The studies presented in chapter 3 - investigating the possibilities of the technique of electrostatic force microscopy (EFM) for carbon nanotube studies - may seem somewhat unrelated to those of the first chapter on electron transport in InAs nanowires (except for both of them belonging to the general field of nano-science). The reason for this separation is to be found in the circumstances and practical problems encountered during the course of the project. During the start-up phase, no nanowires were available for the study of transport and a considerable amount of time was spent dealing with practical issues concerning the infrastructure and construction of a CVD lab for the growth of silicon nanowires. In this period also some efforts were put into optimizing a nanotube CVD setup, and the EFM technique was introduced during the search for the best way of characterizing the nanotube growth products. The various studies in chapter 3 emerged while investigating subtle features of the obtained EFM images. Once semiconducting nanowires were finally grown in our group (enabled by the decision to introduce gold into the existing MBE-chamber) all efforts were directed towards making nanowire devices for electrical transport stud3 ies, and the EFM investigations were paused. After these preliminary comments we can now begin with the presentation of the results of experimental investigations of low-temperature electron transport in semiconductor nanowires. Please enjoy! 4 Chapter 2 Electron Transport in Semiconducting Nanowires 2.1 Introduction During the last 50 years the technological development in the semiconductor industry has been aimed at making electronic devices smaller. At some point the size reaches the scale of the coherence length of the electrons in which case the wave nature of the electron becomes important for the device behavior, leading to a wealth of novel phenomena. This mesoscopic regime has been intensively studied since the 1980ies, in particular utilizing the two-dimensional electron gas of a semiconductor heterostructure (2DEG) as the experimental realization. This was later supplemented by an intense research in mesoscopic devices incorporating carbon nanotubes and other nano structures. This chapter presents our studies of electron transport in one of the most recent classes of mesoscopic systems: semiconducting nanowires (NW). Nanowires, occasionally also denoted nanorods or nanowhiskers, are high aspectratio needle-like crystals typically measuring 2 − 100 nm in diameters and 5 − 30 µm in lengths. During the last few years the research into the properties and possibilities of nanowires has accelerated, leading to many exciting discoveries. This introductory section first provides a short survey of the field of nanowire research followed by a brief introduction to mesoscopic electron transport being at the heart of the work presented in the chapter. The introduction ends with an outline of the chapter. Nanowires In this section we outline the general field and history of nanowire research, postponing, to the subsequent section, the treatment of the properties of the particular type of wires which was used for our measurements: semicon5 Au As In As In In InAs In As Figure 2.1: Top left, schematic illustration of the VLS growth of nanowires (see text). Bottom left, TEM micrographs of Si nanowires grown by laser ablation (adapted from Ref. [1]). Right, SEM micrograph of InAs nanowires grown by CBE from lithographically defined gold particles (adapted from Ref. [2]). ducting Indium Arsenide (InAs) nanowires. The section does not (by any means) cover the complete field which, despite its young age, already encompasses a substantial number of published works. Instead, some general aspects and applications are described being a part of the author’s motivation for entering this field of research. Already in the 1960ies the growth of whisker or wire-like semiconductor crystals were demonstrated[3] and the growth mechanism, the so-called vaporliquid-solid (VLS) growth, is still the governing principle in most presentday nanowire growths1 . The VLS-mechanism is sketched in Fig. 2.1(a): The starting point of the NW growth is a substrate prepared with metal (usually gold) nanoparticles either by deposition from commercially available colloidal suspensions[6], by direct deposition using aerosol techniques[7] or made by metal evaporation and subsequent thermal annealing[8]. Subsequently, the substrate is heated and placed in a gas flux containing the constituents of the desired semiconductor. These are incorporated into the molten nanoparticle thereby forming an eutectic alloy and at the point of supersaturation, crystal growth takes place beneath the particle (rightmost part of the schematic). By examining the phase diagram of the alloying between the metal nanoparticle and the semiconductor species, the growth parameters can be rationally chosen making the VLS mechanism a viable route for the synthesis of a nanowires from wide range of group IV (Si, Ge, SiGe), group III-V (GaAs, GaP, GaAsP, InAs, InP, InAsP) and II-VI (ZnS, 1 Recent work indicates that nanowires in some cases grow by a vapor-solid-solid mechanism[4] and also solution-based techniques have been developed[5]. 6 ZnSe, CdS, CdSe) semiconductors[6, 3, 1]. The diameter of the resulting wire is determined by the size of the metal catalyst particle and the growth of epitaxial wires with diameters as low as 3 nm has been reported[9, 10]. The length is determined primarily by the growth time and can be up to many micrometers. The method used for introducing the semiconductor species varies and nanowire growth has been demonstrated using laser ablation[1], chemical vapor deposition (CVD), metal-organic vapor phase epitaxy (MOVPE)[11], chemical beam epitaxy (CBE)[7], and molecular beam epitaxy (MBE)[12]. The latter method has been employed for growing the wires used in this work (see section 2.2). Figure 2.1 shows examples from the literature of nanowires: The bottom-left images show transmission electron microscope (TEM) images of a Si NW grown by laser ablation[1] and the rightmost panel shows an scanning electron microscopy (SEM) image of InAs nanowires grown by CBE from lithographically defined Au-particles. The nanowires grown by the VLS method possess many unique and exciting properties. The increasing interest in recent years is probably partly spurred by the improvements and better understanding of the growth process enabling rational design of wire geometries and composition. Importantly, in a particular growth, all nanowires are, to a large extend, identical. This contrasts the other famous one-dimensional system, the carbon nanotubes, for which a given growth contains nanotubes of a wide range of crystal structures leading to very different properties (section 3.1). Despite extensive research, this aspect still remains an obstinate problem for the technological application of carbon nanotubes. The VLS growth process furthermore opens up for the possibility of changing the nanowire composition during growth by altering the semiconductor species introduced in the growth chamber. In this way, epitaxial heterostructures, as are known from the two-dimensional growth of semiconductors, can be designed along the wires (axially), as well as radially. In the case of two-dimensional growth the materials which can be epitaxially combined are limited by their lattice mismatch since, for large mismatch, the interface becomes defective due to strain. For the nanowires, however, their small cross-sectional area allows the strain to relax to the surface, enabling the realization of novel material-combinations thereby adding even further flexibility to the nanowire system. The possibilities (fabricating radial and axial heterostructures, diameter control etc.) enable a wide range of functionalities to be introduced in the wires, already at the growth stage, and p-n junctions[13] and grown barriers[14] etc. have been reported. Figure 2.2 shows examples from the literature of nanowire heterostructures. Panel (a) shows a TEM image of a Ge/Si core/shell nanowire with an atomic-resolution zoom shown in (d). In (b),(c) elemental mappings of the Si and Ge in the same view as panel (a) are shown, clearly emphasizing the core/shell structure. This is further elaborated in panel (e) showing the elemental mapping in a cross-section through the wire (adapted from 7 f Figure 2.2: (a) TEM micrograph of a Si/Ge core/shell nanowire with a high resolution zoom in (d). (b),(c) Si and Ge elemental mapping of the image from (a) showing clearly the core/chell structure. This structure is further elaborated on in (e) showing the elemental composition at a cross section of the wire - blue trace Si, red Ge (adapted from Ref. [15]). (f) An TEM image of an InAs nanowire containing InP segments. The image is colored with respect to the lattice spacing (adapted from Ref. [16]). Ref. [15]). Panel (f) shows an image of an InAs nanowire grown with InP segments (barriers). The composition changes on the level of monolayers and the image has been colored according to the lattice spacing (adapted from Ref. [16]). Another characteristic which distinguishes nanowires from conventional nanostructures, which are lithographically defined from two-dimensional substrates, is the possibility of transferring the nanowires from the growth substrate to virtually any other substrate. This enables device fabrication on, for example amorphous SiO2 -wafers suitable for low temperature devices (section 2.3) or plastic films for realizations of flexible electronics[17]. A feature which has received considerable attention is the use of NW’s for sensor applications. The electrical conductivity of semiconducting NW’s is highly sensitive to electrostatic gating and combining this with their very high surface/bulk ratio makes NW’s extremely sensitive to their immediate environment. Much research has been directed towards this application and advanced schemes for single-molecule detection have been realized[18]. As for low temperature investigations, which is the focus of the work presented in the upcoming sections, a number of important results have also 8 been established early on. Resonant tunneling diodes and quantum dots with barriers defined during NW-growth have been realized[19, 14] followed by reports on electrostatically defined single and double quantum dots[20]. Finally, the flexibility offered by nanowire devices concerning the contact material (unlike the strict demands of conventional 2DEG-based mesoscopic devices) has been exploited in Refs. [21, 22, 23] for fabricating nanoscale devices with superconducting contacts. This has enabled the experimental investigation of a number of novel effects such as a tunable[22] and reversible[21] supercurrent. The reported low temperature studies will be further discussed below during the presentation of our own measurements. Mesoscopic transport This chapter describes measurements of the electrical resistance (or conductance) of nanowires. For a macroscopic bulk sample, the electrical resistance is simply proportional to its length and inversely proportional to its cross sectional area, the coefficient of proportionality being the resistivity. Thus, a sample of half the width will simply have twice the resistance but brings no new physical phenomena. This is the Ohmic regime of charge transport. A furious technological development has for the last decades aimed at downsizing of electronic devices. Obviously, at some point the physical behavior must change since eventually the atomic scale is reached, a scale, which is well known to be governed by entirely different principles. New phenomena are, however, observed long before the atomic scale as the sample size L approaches the scale of the single electron coherence length `ϕ . The coherence length is normally determined by inelastic scattering e.g., scattering by lattice vibrations, and `ϕ can be long at low temperatures. In this mesoscopic regime (L ¿ `ϕ ) effects associated with the wave nature of the electrons become important leading to a wealth of new physical phenomena. Several approaches exist for experimentally realizing mesoscopic devices. An important class of systems utilizes the 2DEG which can exist at an epitaxial grown interface between two semiconductors. Employing lithographic techniques to define electrostatic gates the 2DEG can be shaped into advanced geometries allowing a large variety of devices to be realized. Another important class of mesoscopic devices are provided by carbon nanotubes which are less flexible with regard to device design but, on the other hand, are ”born” as nearly ideal one-dimensional conductors. The nanowire devices, which are treated in this chapter, also belong to the mesoscopic regime. There is of course common phenomena associated with the nanowires and the aforementioned systems but (as discussed in the previous section) they also provide novel opportunities as well as unique (experimental) difficulties. Figure 2.3 shows a 3D rendering of an atomic force microscopy (AFM) measurement of a nanowire device. Electrical contacts are made by evaporating 9 Metal electrode Γ2 ee- Insulating substrate Γ1 Metal electrode Nanowire Figure 2.3: A perspective view of an AFM measurement, schematically illustrating the system under consideration: a nanowire with metal electrodes. The coupling of the wire to the leads is characterized by the tunnel rate Γ = Γ1 + Γ2 . metal on the ends of the nanowire2 allowing investigations of the current flow through the wire while varying external parameters. To describe how well the wire and the metal leads are coupled we use the electron tunnel rate Γ as indicated on the figure. Another parameter which is important for the device behavior is the so-called charging energy EC which denotes the electrostatic energy required to add an electron to the wire3 . One of the features of the nanowire system has been the possibility of realizing devices with different barrier transparencies Γ, either electrically by means of a nearby electrostatically coupled metal electrode (gate) or by varying the fabrication techniques. This aspect is illustrated in Fig. 2.4 which shows the linear conductance G (inverse resistance) as a function of the voltage Vg on a gate electrode situated below the insulating top-layer as shown in the device schematic (inset). Obviously, the behavior of the device is qualitatively different in the three Vg regimes. The reason is that the coupling Γ depends on the back gate potential, and as Vg is sweeped the device thus enters different regimes dominated by different physical phenomena. The measurements will be treated in subsequent sections but the regimes can be distinguished by the relationship between Γ and EC [24, 25]: • Γ ¿ EC . In this weak coupling regime corresponding to the lowest Vg region of Fig. 2.4, an electron entering the device spends a long time on the nanowire before leaving. The charge on the wire is a well-defined integer multiple of the elementary charge e, and charging effects dominate. The transport is thus governed by Coulomb blockade 2 3 The fabrication will be described in section 2.3 See section 2.4 10 Vsd 3 Contact NW I G [e2/h] SiO2 2 1 0 -9.0 p++ Si Vg x20 -8.8 -8.6 -4.4 -4.0 -3.6 6 8 10 Vg [V] Figure 2.4: The linear conductance G of a nanowire device at T = 300 mK as a function of the voltage Vg on a capacitively coupled back-gate situated beneath the device (see schematic, inset). The device transparency increases with the gate voltage and three transport regimes can be identified; closed dot (lowest gate region), intermediate coupling (middle gate region) and open device (highest Vg region), respectively. physics. This is the subject of section 2.4. • Γ . EC (middle region of Fig. 2.4). Coulomb blockade is still important in this intermediate coupling regime but higher order tunneling processes contribute significantly to the transport. Cotunneling and the so-called Kondo effect are observed (section 2.5). • Γ À EC (highest Vg region of Fig. 2.4). In this strong-coupling regime the charge on the nanowire is no-longer well-defined and the transport is dominated by effects of interference of electron waves (conductance fluctuations) as treated in section 2.6. Other characteristic parameters relevant in describing the various regimes will be discussed in the subsequent sections in relation to the measurements. Organization of this chapter The outline of this chapter is as follows: After this introductory description of nanowires and mesoscopic transport, the particular nanowire material used in this work is presented. This is followed by a section on the experimental techniques used for device fabrication and electrical measurements at low temperatures (section 2.3). The subsequent presentation of the results is given in sections 2.4, 2.5, and 2.6 which are organized according to the three transport regimes listed above; closed-dot, intermediate coupling, and open dot, respectively. 11 Further interesting physical phenomena are anticipated for mesoscopic devices with superconducting leads and this is the subject of section 2.7. Initially, the additional experimental efforts required for attaching superconducting Al electrodes to the nanowires are described, followed by a presentation of the results. Again, these are organized according to increasing barrier transparency with emphasis on the effect of the superconductors in the regime of intermediate coupling (Kondo effect). 12 2.2 InAs nanowires In the previous section a brief introduction to the general field of nanowire research was provided illustrating that a large number of semiconductor nanowires can be grown. In this section we focus on the properties of the particular nanowires used in this work. Actually, two types of nanowires have been investigated: Indium Arsenide (InAs) nanowires grown by MBE at University of Copenhagen by Martin Aagesen and Claus Sørensen (see details below) and Silicon (Si) nanowires provided by Jiwoong Park at the Rowland Institute, Harvard University. Due to obstinate difficulties in making electrical contacts to the Si nanowires this work was abandoned/postponed when the InAs nanowires became available. The InAs wires are easier to contact as will be discussed below but pose other experimental problems (section 2.3). In this section the InAs nanowire material is introduced and some of their relevant properties are discussed. The nanowire material As mentioned the InAs nanowires were grown by molecular beam epitaxy following the vapor-liquid-solid method described in the previous section (Fig. 2.1). These are the first InAs wires grown by MBE4 and the details on the growth are provided elsewhere[27, 28]. It follows along these lines: A GaAs substrate is loaded in the MBE chamber which is held at a background pressure of ∼ 10−11 torr. The substrate is heated to ∼ 650 ◦ C and ∼ 1 nm gold is deposited in situ. Due to the elevated substrate temperature the gold forms nanosize droplets on the surface which will act as growth promoters for nanowires. During the whole process the sample sits in a flux of As and following the Au deposition the temperature is lowered to ∼ 480 ◦ C and the nanowire growth is started by a few nanometers of GaAs by turning on a Ga source5 . Subsequently the temperature is further lowered to ∼ 400 ◦ C and by opening an In source the growth is continued as an InAs NW. It turns out that the nanowires are epitaxially oriented with respect to the substrate and that they grow in the <111> direction. Figure 2.5(a) shows 4 InAs wires reported in the literature are most often grown by CBE. This technique differs most significantly from the MBE technique in the way the semiconductor compounds are handled. In CBE, the In and As come in the form of metallo-organic molecules (e.g. trimethylindium for the In) whereas in MBE they come in pure form. Thus we may expect that the MBE method introduces less carbon doping ([26]), however, no conclusions can be drawn on the basis of the present early stage of the investigations. 5 Starting the NW growth by a small section of GaAs is required due to the lattice mismatch between InAs and the GaAs growth substrate. Once the nanowire growth has started and the materials are changed to InAs the NW growth continues since the strain can be relaxed to the surface. This step is not needed for growth of InAs NW’s on the (more expensive) InAs substrates. The small section of GaAs in the end of the nanowire is not important for the electrical measurements since the device is patterned in the middle section of the wire. 13 (a) (b) (d) (c) Figure 2.5: (a) Picture of a GaAs substrate after MBE growth of InAs nanowires. The substrate assumes a grey color after the wire growth. (b) SEM micrograph showing a side view of the nanowire forest close to a cleaved edge. The substrate is (100) GaAs and as the wires grow in the <111> directions, they do not stand perpendicular to the substrate as in (c) showing the result of growth on a (111) substrate. (d) Inset: top view of a growth on a (111) substrate; the wires have hexagonal cross-sections epitaxially oriented to the substrate. Upon tilting the sample the SEM image (main panel) shows directly the faceted nanowire. (SEM images courtesy of Martin Aagesen). 14 a picture of a wafer after growth6 . the and in (b) a side-view SEM image shows the nanowire ”forest” close to a cleaved edge of the wafer. In panel (b) the substrate is (100)-GaAs and thus the nanowires do not grow perpendicularly to the surface but form angles consistent with the <111> direction. Indeed, panel (c) shows the result of wires grown on a (111) GaAs substrate where all wires stand in the vertical direction. In the inset to panel (d) such a growth is viewed from the top and the wires can be seen to have a hexagonal cross-section and all wires are oriented the same way confirming that they keep their epitaxial orientation during growth7 . Upon tilting the sample the facets of the nanowire are clearly observed. Figure 2.6 shows high-resolution TEM micrographs of MBE grown InAs nanowires8 . Firstly, from panels (a) and (c) it is noted that the wires are covered with a few nanometers (. 5 nm) of amorphous (oxide) coating. This shows that an oxide-removing treatment is needed to achieve electrical contact to a nanowire (section 2.3). Secondly, the wire obviously contains a number of crystal faults. Consider first the image in (a) and the zoom in (b). From detailed analysis of the image it is found that this nanowire has cubic zinc blende (zb) crystal structure like bulk InAs crystals. The nanowire grows in the <111> direction and contains multiple defects which can be identified as twinning faults, i.e., at certain points the stacking suddenly changes from ABCABC to CBACBA (mirror symmetry). Twinning has been studied in details in Ref. [29] for InP nanowires where it is shown to facilitate micro-faceting of the wire surface. It is uncertain, however, how the defects will affect electron transport. Figure 2.6(c) shows a TEM micrograph of a different InAs nanowire: Again segments of twin-related zb structures are observed (labeled a1 , a2 ) but large sections of the crystal (labeled b) are found to have a hexagonal wurtzite (wz ) structure (i.e. ABAB stacking) which is not found in bulk InAs. The wz structure has been observed in a number of nanowire studies (see e.g. Ref. [30, 31]) but very little is know about the properties of this phase. Two recent theoretical works (Ref. [32, 33]) report on band-structure calculations and estimate a ∼ 13% (55 meV) increase in the band gap of the wz structure, compared to the band gap of zb InAs Egzb = 0.43 eV (at T = 0 K,[34]). Thus, we may expect the wz nanowires to behave slightly different and abrupt changes between wz and zb structures in the same wire, as observed in Fig. 2.6(c), to influence electron transport. The calculations do not, however, include the finite size of the NW crystals which may also affect the band structure significantly. 6 The wire forest causes the greyish color of the central area of the wafer, and the darker and more shiny appearance of the last few millimeter closest to the edge corresponds to the naked substrate. No wires are grown close to the edge due to the clamping of the wafter during growth. 7 There are exceptions to this, and in some growths, wires with different directions may be present. For further discussions see Refs. [27, 28]. 8 Imaging and analysis courtesy of Erik Johnson, University of Copenhagen 15 (a) (b) <111> (c) 10 nm 0001h b 0110h 2110h a1 b 111c 211c a2 110c a1 Figure 2.6: TEM micrographs of InAs nanowires. (a) Low-resolution image showing a ∼ 40 nm diameter nanowire of cubic zinc blende structure containing multiple twinning defects. The zoom in (b) illustrates the mirror symmetry in the planes of the twin defects. (c) TEM of a nanowire having regions of zinc blende structure (labeled a) and regions of hexagonal wurtzite structure (labeled b). Only the zinc blende structure is found in bulk InAs crystals. (Images and analysis courtesy of Erik Johnson). More work is needed to understand these issues. All measurements presented in this chapter were carried out on devices fabricated from the InAs nanowire growth #NBI73 using a (100) GaAs substrate. The growth contains wires of various lengths and diameters but those which were chosen for transport studies have diameters ∼ 60 nm, lengths up to 4 µm and most show virtually no tapering. SEM images from this particular growth are provided in Appendix A. Electronic properties of InAs nanowires Due to the lack of knowledge about the properties of the wz structure, we briefly mention here some of the relevant well-established properties of bulk zb InAs. As we will see later, the measurements are to a large extend consistent with these properties (and effects of confinement). Bulk InAs has a small, direct band gap of Egzb = 0.43 eV and an low effective mass 0.026 me [34]. An important property of InAs for our purpose is that it 16 makes a Schottky barrier free contact when brought in contact with Au (and Ti) metals due to surface states which pins the Fermi level of the metal in the conduction band[35]. Thus, contacts are established by simply evaporating metal onto the wires, in contrast to e.g. Si NW, for which carefully controlled thermal alloying is required to avoid the otherwise all-dominating Schottky barriers at the interfaces. Furthermore, bulk InAs has a large g-factor |gbulk | = 15 leading to a large Zeeman splitting of the energy levels in a magnetic field (section 2.5). The g-factor in nanowires is, however, suppressed with respect to the bulk value due to the geometric confinement[30]. Finally, the bulk crystal exhibits a strong spin-orbit coupling being responsible for the presence of weak antilocalization in magneto-transport measurements[36, 37] (section 2.6). 17 2.3 Device preparation and experimental details In the previous section the synthesis and properties of semiconducting InAs nanowires were described. As shown in the SEM image of Fig. 2.5 the nanowires are grown as a dense nanowire forest on a substrate. This section describes the methods that were developed to fabricate devices enabling electrical measurements on individual nanowires from the forest. The methods overlap considerably with the techniques used for fabrication of carbon nanotube devices, however, the nanowires do pose their own obstacles which need special attention. In the final part of the section the methods used for performing low-temperature measurements are briefly described. Device fabrication The goal of this section is to briefly describe the methods used to achieve macroscopic electrical contacts to nanowires in a simple field-effect-transistor (FET) geometry shown schematically in Fig. 2.7(c) with two electrical contacts and the possibility of electrostatic gating by a global back-gate electrode. A detailed account of the fabrication scheme, including various parameters, are given in Appendix B. The fabrication follows along these lines: • The starting-point is substrate wafer of highly doped silicon capped with an insulating layer of SiO2 . The wafer is prepared with alignment marks and bonding pads by lithographic techniques. • Wires are transferred from the growth substrate (Fig. 2.5(b)) by gently pressing the two wafers together. In this way the longest wires break of the growth substrate and sticks to the silicon wafer. • Wires are located with respect to the alignment marks by SEM or optical microscopy, and an appropriate contact pattern is designed. Fig. 2.7(a) shows a typical optical microscopy image of nanowires deposited in a grid of metal alignment marks. • Contacts were defined by e-beam lithography. Prior to metal evaporation, the samples were cleaned in an oxygen plasma etch to remove resist residues from the wire surface. This was followed by a brief wet etch in buffered hydrofluoric acid (HF) to remove the oxide from the nanowire surface (cf. TEM image, Fig. 2.6). In order to limit re-oxidation of the wire, efforts were made to bring down the time between the end of the HF etch and loading in the evaporation chamber. Usually the load-lock was evacuated within 2 minutes of the HF etch. Based on the relatively few devices (< 20) which were electrically characterized it seems that omitting the plasma and HF etches result in devices with large contact barriers (devices in the Coulomb blockade regime). In contrast, by including the cleaning procedure, 18 (a) Connection to bonding pads Alignment marks Nanowire 20 µm (c) (b) Ti/Au contact Nanowire Au Ti Nanowire SiO2 doped Si Figure 2.7: (a) Optical microscope image showing a grid of alignment marks fabricated on a Si wafer. The marks are used as references when designing contact patters for nanowire devices. Part of the larger structure which allows connection to macroscopic bond pads is also seen. Dark lines corresponds to InAs nanowires which have been deposited on the sample after the fabrication of the alignment grid. (b) An SEM micrograph of a typical nanowire device. Metal Ti/Au contacts have been defined with an electrode spacing of ∼400 nm. (c) Schematic side-view of the device layout of (b). 19 better contact is achieved and the device transparency can be tuned by the gate (section 2.1). For the contacts either a Ti/Au bilayer was used (for normal-metal contacts) or a Ti/Al/Ti trilayer for superconducting contacts (see also section 2.7). Figure 2.7(b) shows an SEM micrograph of a typical finished device. • An ultrasonic ball bonder was used for bonding devices, however, special care was needed since the nanowire devices were found to be extremely sensitive to electrical/static shocks. Hundreds of devices were destroyed in the bonding process and the subsequent handling. A number of precautions enabled, however, a reasonable fraction of devices to reach the measurement step (see Appendix B for details). Most importantly it was found that after bonding each device the sample should be mounted in the cryostat and checked for conduction. In case the device survived the bonding it was cooled and investigated at low temperature. Attempting to bond additional devices on the same chip usually results in the loss of all previously bonded devices. This procedure greatly limits the throughput, but no other solution has yet been found. Electrical measurements This section briefly describes the setups used for performing electrical measurement on the nanowire devices. Almost all the measurements were carried out in a voltage-biased setup using lock-in techniques as schematically shown in the diagram of Fig. 2.8. The measurements were controlled by Lab-View software running on a PC, equipped with a National Instruments data acquisition (DAQ) card (DA and AD converters). To reduce noise, the sample was isolated from the mains by battery driven opto-couplers (isolation amplifiers with unity gain) on all lines connecting to the sample. The output range of the DAQ is ±10 V and one output (DAC1 on Fig. 2.8) controls directly the gate potential through a large 10 MΩ resistor which protects the sample in case of a gate-leak. The source-drain bias is supplied through DAC0 but since we are usually interested in biases in the milli-Volt range, the output is passed through a 1000:1 voltage divider before it is connected to the sample. On top of the DC bias a small AC signal is added for lock-in measurement. The current is measured using a low-noise battery powered current amplifier (DL Instruments, model 1211) which outputs a voltage proportional to the current. This output (DC current) is logged by ADC0 and is fed into the lock-in amplifier which measures the AC part of the current (at the same frequency as the applied AC bias). Finally, the output from the lock-in, corresponding to the differential conductance dI/dV of the device, is logged by ADC1. Initially measurements were carried out in a simple dip-stick setup in which 20 Sample Current amp. I DAC0 1000:1 (DC bias Vsd) 10000:1 10MΩ (AC source) ADC0 (DC current I) DAC1 (gate Vg) Opto coupler Ref. in ADC1 (dI/dV) Lock-in amp. Figure 2.8: Schematic diagram of the electrical setup used in the measurements. A computer controls/sweeps the DC bias and gate potential and measures the DC current as well as the differential conductance which is extracted using a standard AC lock-in setup. the sample can be lowered gradually into a transport dewar with liquid helium. At different distances from the He-level the sample reaches equilibrium at different temperatures, thus enabling measurements from room temperature to 4.2 K. It turned out, however, that the phenomena of interest in this project required even lower temperatures and thus the samples needed to be transferred to a more advanced setup (described below). The dip-stick setup is, however, usually a very fast and convenient method for screening samples for the best devices which is then subsequently cooled to lower temperatures. For the nanowires, however, once the sample is mounted in the dip stick and the bonding-wire protection loops removed to enable measurements (see above) it has proven exceedingly problematic to transfer the sample to another setup without loosing the device9 . Thus, after a while the otherwise convenient dip-stick pre-screening was abandoned and the samples were directly mounted in an Oxford Instruments Heliox system. In the Heliox the sample sits in vacuum and the whole inset is submerged into liquid He. By pumping on liquid 4 He, present in a small can in the Heliox inset, a temperature of ∼ 1.3 K is reached. This is sufficient for condensation of 3 He of which the Heliox holds a small amount. Due to the higher vapor pressure of 3 He (for a given temperature) even lower temperatures are reached by subsequently pumping on the condensed 3 He liquid. The sample temperature in this setup reaches, in this way, a temperature of ∼300 mK and the hold-time is roughly 24 hours. For further details on 9 This problem is obviously not a fundamental one and a scheme has subsequently been developed to keep the sample properly grounded at all times while transferring. 21 (a) (b) (c) (d) 1µm Figure 2.9: (a)-(d) A series of successive AFM images showing the effects of AFM manipulation. The wires are relatively easily moved around on the surface and the figure illustrates the possibility of assembling advanced nanowire devices geometries, e.g., using AFM manipulation. the principle and operation of the relatively simple Heliox setup, the reader is referred to the manual and Ref. [38]. The setup is equipped with a 2 T superconducting magnet and is placed in a shielded room. Advanced device geometries We conclude this section by commenting on the possibility of assembling the InAs nanowires into advanced device geometries. A number of device architectures incorporating multiple silicon nanowires have been developed for sensor purposes[18] and for making junctions between wires with different properties in a crossed-wire geometry[39]. In these studies the wires were assembled either by methods of directed flow or by direct mechanical micro/nano manipulation in an optical microscope. These methods benefit from the extended length (> 10 µm) of these silicon wires and may not be applicable to the shorter MBE-grown wires. Figure 2.9 demonstrates the use of AFM manipulation for moving individual InAs nanowires which were deposited on a silicon substrate as described above. The wires are found to be robust towards pushing by the AFM (in contrast to single-wall carbon nanotubes, see section 3.5) and they are, with relative ease, assembled into a crossed-wire geometry in four manipulation steps. Future studies may benefit from this possibility. 22 2.4 Nanowire devices in the closed dot regime In this section we begin the presentation of the low temperature electron transport measurements performed on the devices described in the previous section. The results are presented in the order of increasing barrier transparency starting with this section on nanowire devices with large barriers. In this regime, the device acts as a quantum dot and transport is dominated by the phenomenon called Coulomb blockade (CB). Coulomb blockade physics has been observed in a large number of mesoscopic devices and the phenomenon is well understood. Thus, the observation of CB in the nanowire measurements comes as no surprise, however, these measurements are the first performed on the ”Copenhagen-wires”, which differ from other wires in that the growth is performed by MBE and, as shown below, that the barriers are defined by the contacts. Thus, the CB measurements add to the understanding of these wire’s ability to act as mesoscopic devices and the results are therefore relevant for this work. Furthermore, a thorough understanding of the CB in these simple devices is a prerequisite (or at least an aid) for future development of advanced nanowire devices. Lastly, this section introduces the terminology and sets the scene for the following sections describing the results in the regimes of stronger coupling. Devices in the closed dot regime have been realized using two fabrication schemes differing in the procedure used for ”cleaning” the wire interface before metal evaporation (section 2.3). As mentioned in section 2.1, exposing the wire to both oxygen plasma etching and a HF etch prior to metal evaporation, results in devices where the barriers are tunable with the back gate (type-1). In this case (see Fig. 2.4) the closed-dot regime is relevant for the lowest gate-voltages when the barriers are large. Alternatively, devices born with large barriers can be fabricated by omitting one of the cleaning steps. Here we show results from devices fabricated without the plasma etching (type-2). The type-1 devices are generally regular and stable in the intermediate coupling regime but in the closed-dot regime, relevant for this section, they become noisy and exhibit multiple gate-switches. Therefore, regular Coulomb blockade was only observed in relatively short gate-ranges. The type-2 devices, on the other hand, are usually stable and regular in the closed-dot regime but it has not been possible to tune these devices into the intermediate and open regimes. Figure 2.10 shows the linear conductance G as a function of gate-voltage Vg for various temperatures T between room temperature and 4.2 K. The leftmost inset shows an SEM micrograph of the device. The contacted wire has a diameter d ≈ 70 nm and the (Ti/Au) electrodes are separated by L ≈ 300 nm. Importantly, for all temperatures the conductance increases upon increasing Vg showing that the current is carried by electrons; i.e., the nanowire acts as an n-type semiconductor. The n-type behavior has been observed for all investigated InAs nanowire devices agreeing with the 23 G [e2/h] 2.0 G [10-3 e2/h] 2.5 295K 1.5 6 4.2K 4 13.0 0.25 1µm 13.3 Gate [V] 13.6 210K 100K 0.00 50K 4.2K -10 0 10 20 Gate [V] Figure 2.10: The linear conductance G as a function of gate voltage Vg measured at various temperature from room temperature to 4.2 K for a type2 nanowire device with a ∼ 70 nm diameter wire and ∼ 300 nm electrode separation (SEM micrograph, leftmost inset). The rightmost inset shows a magnification of the T = 4.2K trace in the region marked with a red oval in the main panel. The oscillatory behavior is due to Coulomb blockade. These characteristics are typical of all devices. measurements on CBE grown InAs nanowires of Ref. [26] where the carbon present during the CBE-growth was speculated to be the source of the donors. In our case, however, there is no carbon present during the (MBE) growth but the wires may be subsequently doped by the environment or the n-type behavior may be an effect of the peculiar contact properties of InAs (section 2.2). Another feature of the measurements in Fig. 2.10 is the decreasing of conductance observed upon lowering the temperature which shows the importance of thermal excitation of carriers into the conduction band of the wire. The rightmost inset shows a blow-up of the T = 4.2 K trace at the region marked in the main panel with a red oval. The conductance exhibits pronounced periodic oscillations which are caused by Coulomb blockade as described below. Coulomb blockade - model In the following, the nanowire device is modeled as a conducting island (dot) coupled by tunnel-contacts to the two metal electrodes. For poorly coupled devices G < e2 /h the charge on the island is a well-defined integer number N of the elementary charge e. If each tunnel junction is represented by a capacitance and a resistance the model can be illustrated by the circuit in Fig. 2.11, where the back gate electrode is included as a capacitive coupling Cg . If we assume that the total capacitance C = Cg + Cs + Cd accounts 24 Rd Nanowire Rs N Cd Cs Cg Vg Vsd Figure 2.11: Circuit diagram schematic of the nanowire device. Each wire/electrode junction is represented by a resistance and a capacitance and the gate couples capacitively to the central region. For high-resistance devices the charge on the nanowire-dot is a well-defined integer number N of electron charges. for the Coulomb interaction of the electrons on the island, the change in electrochemical potential associated with the addition of an extra charge on the dot is the charging energy EC = e2 /C independent of N . Obviously, this energy is large for small objects and for low temperature measurements where kB Tk < EC effects associated with single electron transport through the device become important. Single electron transport has been thoroughly analyzed in the reviews of Refs. [40, 41] and the details will not be repeated here. The treatment in this section serves to describe the physical origin of the oscillatory behavior of G observed in Fig. 2.10, and to introduce the relevant energy scales. In the model of Fig. 2.11 the reduced size of the nanowire is not evident. However, due to the quantum nature of the electrons, the nanoscale confinement of the nanowire is expected to lead to a discrete set of electron energy levels {En }. These are important for the measured transport characteristics if the experiment is carried out with kB T < ∆En , where ∆En = En − En−1 are the spacings between the successive single electron levels. In this case the island is denoted a quantum dot. An alternative view in which these discrete energy levels are included is provided by the sketch in Fig. 2.12(a) of the potential landscape through the device: Tunnel barriers separate the dot (quantum well) in the center from the source and drain electrodes which are kept at the chemical potentials µs and µd , respectively. A source-drain bias Vsd = (µs − µd )/e is applied across the dot. The single electron levels are illustrated by horizontal lines; solid for the occupied and dashed for the unoccupied. We assume in the following that the level structure is independent of N . The electrochemical potential of the dot is calculated from the total ground 25 (a) (b) (c) .. ... ∆Ei Eadd e µs Eadd µd µs µd µs Eadd Vsd µd .. .. . .. . .. Vsd e Vsd Figure 2.12: Schematic potential landscape of the nanowire device. The electrode/wire junctions are represented by tunnel barriers and a small bias Vsd = (µs − µd )/e is applied across the device. Due to the nanoscale confinement of the wire (central region) the electrons occupy discrete energy levels illustrated by horizontal solid(dashed) lines for occupied(unoccupied) levels. In (a) there are no levels for transport in the bias window and transport is blocked. The gate voltage displaces the level structure vertically and in (b) a level has been tuned into the bias window and a current can flow by single electron tunnel events on (b) and off (c) the dot (see text). state energy µN = U (N − 1) − U (N ), where U contains a contribution P N Ei due to the discrete energy levels and an electrostatic contribution 2 Q /2C = (e(N − N0 ) + Cg Vg )2 /2C with N0 being the number of electrons at zero gate voltage. This leads to[40] µN = EN + 1´ e2 ³ N − N0 − − eαVg , C 2 (2.1) where the coupling of the gate is described by α = Cg /C. Thus, the addition energy Eadd depicted in Fig. 2.12(a) is given by Eadd = µN +1 − µN = e2 /C + ∆EN . In the situation illustrated in Fig. 2.12(a) there are no levels available for transport within the bias window. Thus, transport through the device is blocked and the number of electrons on the dot is fixed; this is known as Coulomb blockade. As seen from Eq. 2.1, however, the gate potential Vg continuously changes the energy of the electrons on the dot and thereby displaces the level structure of the dot-region in Fig. 2.12(a) in the vertical direction. Thus, Coulomb blockade can be lifted by sweeping Vg until the situation of Fig. 2.12(b) is reached. Here (say) µN now falls in the bias window and an electron can tunnel onto the dot leading to the situation of Fig. 2.12(c) with N electrons on the dot. This N th electron can, however, leave the dot again by tunneling to the drain thereby returning the system to the situation of 2.12(b). Obviously, the cycle can be repeated and by such successive single electron tunnel events, a current can flow through the device. Upon further changing the gate voltage, the levels are tuned out of 26 (a) (b) (c) ∆Ei ∆Ei µs µd µs Vsd .. µd µd Eadd Eadd Eadd Vsd ∆Ei e µs e .. Vsd .. Figure 2.13: Lifting of Coulomb blockade by increasing the bias. (a) No current flows at low bias due to Coulomb blockade. By increasing Vsd a level becomes accessible in the bias window (b) and transport occur by a cycle analogous to Fig. 2.12(b),(c). Upon further increasing Vsd the first excited state enters the bias window resulting in a further increase of the current (c). the bias window leading again to a blocked situation analogous to the one in Fig. 2.12(a), however, with the electron number now fixed at N + 1. Thus, from Eq. 2.1 it is seen that upon sweeping Vg , conductance peaks (Coulomb peaks) will appear spaced by ∆Vg = Eadd /(eα), (2.2) and this is the origin of the oscillatory behavior observed in the measurements of Fig. 2.10. The line shape of the Coulomb peaks depends on the coupling strength and the temperature. The lifetime broadening hΓ gives the peaks a Lorentzian line shape with a finite width even at the lowest temperatures and for larger temperatures the width of the Fermi distribution of the electrons in the source and drain contacts results in thermal broadening of the CB peaks. In the quantum limit hΓ ¿ kB T ¿ ∆E, EC the line shape of the peaks are given (for a peak at zero gate) by G(Vg ) ∝ ³ αeV ´ 1 g cosh−2 , kB T 2kB T (2.3) which has a full-width-at-half-max of eα∆VgF W HM = 3.5kB T and a peak height which scales as T −1 . In the classical regime ∆E < kB T < EC the peak shape is similar, but the width scales as ∼ 4.4kB T /(eα) and the peak height becomes temperature independent [42]. The temperature broadening is the cause of the large width of the conductance peaks in the measurement of Fig. 2.10 as well as the non-zero conductance in the valleys (see below). In the above treatment only the linear response limit (Vsd ¿ ∆E, EC ) was considered, and CB peaks appeared whenever a level was tuned by the gate 27 Vsd (c) (d) (e) (f) (h) (g) (f) (b) (c) (d) (e) (h) (g) (f) (e) (a) (b) (c) (d) N (g) (f) (e) (d) (a) (b) (c) (f) (e) (d) (c) (a) (b) (c) (d) ∆Ecd Eadd ∆Ecd N-1 (f) (e) (d) (c) N N+1 Eadd (e) (f) (g) (h) . eαVg Vsd (f) (g) (h) (e) (d) (c) (b) (e) (f) (g) (h) (d) (c) (b) (a) (d) (e) (f) (g) (c) (b) (a) . (c) (d) (e) (f) Figure 2.14: (Left) Schematic illustration of Coulomb diamonds and the excited state structure in the differential conductance of a quantum dot as a function of bias and normalized gate. The states are labeled according to the schematic on the right which shows the configuration for a point (Vsd , αeVg ) in the center of the middle diamond. The small arrows indicate the level spacing ∆Ecd also indicated in the schematic on the right. into the (small) bias window. The sketches in Fig. 2.12 show that this situation can also be reached upon increasing the bias. This is illustrated in Fig. 2.13: in (a) transport is prohibited due to CB but in (b) it is seen how current can flow at higher bias. In the sketch in (b) the positions of the levels with respect to the drain are the same as in (a). This requires a compensation by the gate, since the levels in the dot are dragged along, when increasing the bias due to the capacitive coupling of the contacts (Fig. 2.11). It is clear in (b) that by moving the levels slightly up with the gate potential the device would re-enter the CB situation and a slightly larger bias would be needed to reestablish transport. Thus, in a plot of the current, or dI/dVsd as a function of Vg and Vsd , the pairs of (Vg , Vsd ), where transport is blocked, form diamond shaped regions. Outside the diamonds features are expected each time an excited state becomes accessible in the transport window as illustrated in panel (c). Such a plot is called a stability diagram or a bias spectroscopy plot, and is schematically illustrated in Fig. 2.14, where different levels can be tracked. Below, these considerations will be used for analyzing the nanowire data. InAs nanowire quantum dots After these general considerations of the transport properties of quantum dots, we now return to the measurements on the InAs nanowire devices. In Fig. 2.10 the overall temperature dependence of the linear conductance 28 was shown and at 4.2 K an oscillatory behavior was observed. From the above discussion, these oscillations are interpreted at temperature broadened Coulomb oscillations. Figure 2.15(a) shows an analogous measurement at a lower temperature T = 300 mK. The measurement is performed in a different cool-down and the transport onset has been shifted to a somewhat lower gate voltage probably due to rearrangements of the charges in the substrate10 . At low Vg no current flows but above ∼ 3.5 V a series of pronounced peaks separated by regions of zero conduction are observed, being clear indications of Coulomb blockade. The peak spacings are regular over the entire gate region spanning ∼ 150 peaks. Superimposed on the fluctuating peak amplitudes an overall gate dependent modulation is observed and they are almost entirely suppressed in the middle region (see below). At the highest gate voltages this modulation causes a stronger coupling to the wire and the conductance does not fall to zero in the CB valleys. The present section focuses on the regular, weakly coupled regions. Panel (b) shows a blow-up of 6 CB peaks measured at 4 different temperatures. As expected from the previous section the width of the CB peaks increases with temperature and in panel (c) the FWHM, ∆VgF W HM , is shown as a function of temperature for the 6 peaks. The dashed line shows the linear dependence of Eq. 2.3 allowing an estimate of the gate coupling coefficient α = Cg /C ≈ 0.086. This shows that the capacitance of the dot is primarily to the contacts which is intuitively reasonably given the geometry of the device (if the device acts as a classical dot, cf. the previous section, the fit corresponds to α = 0.068). As for the peak heights, panel (d) shows the temperature dependence for the peaks in (b): the heights increase as expected for a quantum dot and scales in reasonable agreement with the expected T −1 dependence. Figure 2.15(e) shows a bias spectroscopy plot showing 58 regular low conductance Coulomb diamonds measured in the low Vg region of panel (a). The overall modulation of the contact is observed as regions of increased conductances appear in the vicinity of Vg ∼ 3.9 V and Vg ∼ 5.1 V corresponding to the larger peak amplitudes in panel (a) at these gate voltages11 . For a dot in the quantum regime, the Coulomb peak amplitudes are expected to fluctuate randomly between peaks due to transport through differently coupled levels. In Fig. 2.15(a) an obvious correlation is observed between the amplitudes of adjacent peaks but as seen from Fig. 2.15(e) this appears to be related to the overall modulation of the contact. The origin of this modulating behavior is not clear but it has been observed for all devices, however, 10 Such large differences in the threshold potential between successive cool-downs are frequently observed. Moreover, often a hysteretic behavior is observed upon reversing the sweep-direction of Vg , showing the dependence on the particular charge configuration in the substrate. 11 A small gate shift has occurred in the time between the measurements (days). 29 (a) G [10-2 e2/h] 1.5 1.0 0.5 0.0 4 6 10 8 Gate [V] 0.30K 0.65K 1.00K 1.60K 0.4 0.2 0.0 4.1 4.0 2.0 0.0 0.0 0.5 1.0 1.5 4.2 Vsd [mV] (d) 10-3 Gate [V] 3.0 10-2 6.0 (c) G0 [e2/h] (b) FWHM [mV] G [10-2 e2/h] 0.6 0.3 1.0 T [K] T [K] (e) 0.0 -3.0 3.5 4.0 4.5 5.0 5.5 6.0 Gate [V] Figure 2.15: Measurements of Coulomb blockade physics in InAs nanowires. (a) The linear conductance G as a function of gate volgate measured at 300 mK. At low gate voltages no conduction occurs but above Vg ∼ 3.5 V CB peaks appear. (b) The measurement of 6 CB peaks for temperatures T = 1.6 K, 1.0 K, 0.65 K, 0.30 K; upon lowering the temperature the peaks narrow and the peak values G0 increases as emphasized in panels (c) and (d), respectively. In (c) the dashed lines corresponds to ∆VgF W HM = 3.5kB T /(eα) (Eq. 2.3) with a gate-coupling coefficient α = 0.086. In (d) the solid line corresponds to G0 ∝ T −1 . (e) Measurement at T = 300 mK of dI/dVsd as a function of Vg and Vsd in a gate range corresponding to the left part of panel (a), (darker = less conductive). Coulomb diamonds and large scale conductance modulations are observed (see text). 30 with large differences between devices12 . We speculate that it may be related to charging effects in the electrode-nanowire interface and/or effects of defects in the wire itself close to the interface, cf. section 2.2. Non-linear transport Figure 2.16 focuses on the details of a smaller region of Fig. 2.15(e). Panel (a) shows the linear conductance (left axis) as a function of gate voltage, revealing the regularly spaced CB peaks. The extracted peak spacings (right axis) are grouped around ∼ 40 mV but show a systematic alternating (even/odd) pattern over the entire region. Within the constant interaction model discussed in the previous section, the spacing is given by ∆Vg = Eadd /(eα) (Eq. 2.2) where Eadd = e2 /C + ∆E, and, thus, an alternating behavior is expected in the case of two-fold degenerate levels. Here, the eα∆Vg -sequence would be: (EC , EC + ∆Ei , EC , EC + ∆Ei+1 , . . .). The origin of the degeneracy is not revealed in the present measurement but an obvious candidate is spin-degeneracy which is consistent with measurements discussed below (Fig. 2.17 and Fig. 2.22) of the magnetic field dependence showing that electrons enter the dot in order of alternating spins (↑, ↓, ↑, . . .). In the case of spin-degeneracy being at the root of the alternating ∆Vg the larger spacings in Fig. 2.16(a) correspond to the number of electrons on the dot N being even, while the smaller ones, to N being odd. Thus, from the constant interaction model the smaller spacings should be identically EC /(eα). This is in good agreement with the measurements giving an average charging energy of ECavg /(eα) = 39 ± 1 mV (below, a better estimate of α will be found). The deviations observed on the figure are due to the uncertainty in determining the peak positions or from a slightly varying charging energy. The alternating sequence in principle allows a determination of the subsequent level spacings, however, due to the small level spacings compared to the charging energy and the mentioned uncertainty of the peak positions such a mapping is not meaningful in the present case. The average level spacing ∆Eavg /(eα) = 3.1 ± 1.9 mV, however, gives an idea of the associated scale. Panel (b) shows a stability diagram measured in the gate region marked in (a) with a horizontal dashed line. Five Coulomb diamonds can be clearly seen along with a complicated pattern of closely spaced excited states (cf. Fig. 2.14). The diamonds appear slightly skewed due to an asymmetry in the capacitances of the two barriers13 . From the height of the diamonds we can directly determine the addition energy Eadd /e which should then be linearly related to the peak separations from (a), with a slope corresponding to the gate coupling coefficient. This is confirmed in the inset to panel (a) giving 12 Note that only type 2 devices are considered, cf. the introduction. The asymmetry of the biasing, cf. Fig. 2.8, is not expected to be important since the capacitance to the gate is much smaller than the capacitance to the contacts (α ≈ 0.06). 13 31 (a) 2.0 ∆V [mV] 1.0 1 0 0 0.0 3.8 30 2 20 ∆Vg [mV] 50 10 0 4.0 4.2 Gate [V] 4.4 (b) 5.0 0.01 dI/dV [e2/h] Vsd [mV] ∆Vg [mV] G [10-2 e2/h] 40 3 ∆E/e 0.0 0 -5.0 4.30 4.40 4.35 4.45 Gate [V] Figure 2.16: (a) Linear conductance G (left axis) as a function of Vg showing a series of CB peaks. The peak spacings ∆Vg = Eadd /(αe) (right axis) exhibits an alternating pattern due to a two-fold degeneracy of the levels (see text). (b) Stability diagram for the gate region marked in (a) with a horizontal dashed line (darker = more conductive). Regular coulomb diamonds are observed along with a densely spaced pattern of excited states. The vertical bar illustrates the average level spacing ∆Eavg ≈ 0.18 meV deduced from the even-odd pattern of Eadd in (a) - see text. The inset to (a) shows the diamond width ∆Vg vs. the diamond height in ∆V from (b). The linear fit gives an estimate of the gate coupling coefficient α = 0.06. All measurements were performed at T = 300 mK. 32 α = 0.06 in reasonable agreement with the estimate above based on the temperature dependencies. With this value for α, the average charging energy and level spacing becomes ECavg = 2.3 meV and ∆Eavg = 0.18 ± 0.10 meV, respectively. By comparing with ∆Eavg (vertical bar) it appears that the observed excited states in the stability diagram of Fig. 2.16 do indeed correspond to individual separated levels despite the close spacing of the excited levels and the fact that some appear very faintly. We note that the double degeneracy of the levels identified above should in principle lead to diamonds having pairwise identical excited state spectra. However, it has not been possible to firmly identify such behavior in the stability diagrams in Fig. 2.16(b), due to the large number of closely and evenly spaced levels. Figure 2.17(a) shows a stability diagram measured in the high gate region of Fig. 2.15(a). Again a very regular series of Coulomb blockade diamonds are observed corresponding to addition of 12 electrons. The charging energy EC ≈ 2.3 meV is identical to the previous measurements for lower Vg , however, no even/odd behavior can be identified in Eadd in this region. Except for one strongly coupled level, all lines corresponding to excited states appear with a negative slope. This is in contrast to Fig. 2.16(b) which displays no difference between the two sets of lines (positive/negative slope) and it shows that the dot in the present case has developed asymmetric tunnel couplings to the two leads14 (cf. the above discussion of the overall modulation of the barriers). As in Fig. 2.16(b), the excited states appear closely spaced and identifying the individual level spacings in each diamond is not possible. In contrast to Fig. 2.16(b), however, some of the excited states are much more strongly coupled to the electrodes and the position of these levels can be tracked between neighboring diamonds. The symbols around diamond #3 mark five such states and in panel (b) the threshold biases for each of these have been mapped as a function of diamond number n. As the levels are filled, the positions of the strongly coupled levels shift according to the schematic in Fig. 2.14 and thus, in principle, each set of points in (b) maps the excited state spectrum. As before, however, the relative uncertainty in determining the level positions hinders such detailed analysis. The linear fits to the level positions (dashed lines) are approximately identical (except for a sign) as expected since the same level spacings are responsible for all the position shifts. The average slope yields a good estimate for the average level spacing ∆Eavg = 0.10 meV in reasonable agreement with the (less accurate) value deduced from the even-odd pattern in 2.16(b). We note that the absence of the two-fold (spin) degeneracy in this region is also reflected 14 In the case of asymmetric couplings the transport will be dominated by the thickest barrier and at a finite bias only for those excited states which are accessible for the thick barrier will a line appear in the stability diagram. This can be realized by considering schematics like those of Fig. 2.13. 33 Vsd [mV] 5.0 (a) 2.5 0.0 1 2 4 5 ∆V7 6 8 9 10 11 12 -2.5 -5.0 6.8 6.9 7.0 Gate [V] 7.1 7.2 2.5 (b) (c) 6.95 2.0 57 (d) g=2 ∆V6 ∆V6 1.0 ∆V7 ∆Vg [mV] 7.00 1.5 Gate [V] Vexc [mV] 51 ∆V8 7.05 ∆V7 45 39 ∆V8 ∆V9 0.5 33 7.10 ∆V9 0.0 0 4 8 12 0.0 1.0 B [T] n 2.0 0.0 1.0 B [T] 2.0 Figure 2.17: Grey scale representation of dI/dVsd vs. Vg and Vsd at the high-Vg end of Fig. 2.15(a) showing regular Coulomb diamonds numbered according to increasing electron number n (darker = more conductive). The symbols in diamond #3 label five particularly strongly coupled excited states whose position can be tracked for different n (see Fig. 2.14) and panel (b) shows the threshold bias for each of these states as a function of n. The dashed lines are least square linear fits and the average slope gives an estimate of the average level spacing ∆Eavg = 0.10 meV. Panel (c) shows a grey scale plot of dI/dVsd vs. Vg and perpendicular magnetic field B measured at zero bias for the gate region of diamonds #5-10 (darker = more conductive). The positions of the CB peaks are nearly constant but small wriggles can be observed as emphasized in (d) showing the CB peak separations ∆Vn (three top curves, each offset by 6 mV). For low fields the peaks shift systematically by an amount consistent with a g-factor of ∼ 2 (dashed line, see text). All measurements were performed at T = 300 mK. 34 in level positions of panel (b) which, in the case of spin degeneracy, would only change for every second n. Nanowire dots in magnetic fields Additional information about the nanowire quantum dots can be obtained by applying an external magnetic field B. In the case of spin degeneracy, the degeneracy is lifted due to the Zeeman splitting ∆EZ = 2× 12 gµB B where µB is the Bohr magneton and g the effective g-factor. Thus the level spacings change proportionally to ∆EZ resulting in a shift in the addition energies which can be directly measured by mapping the Coulomb peak separations as in Fig. 2.16. Such measurements were analyzed in details in Ref. [30] for InAs nanowire dots of various sizes showing a variation of the g-factor from the bulk value gbulk = 15 in large dots to ∼ 2 for very small dots. A similar effect is expected also in the absence of spin degeneracy as the mutual distances between the levels shift according to the Zeeman splitting. Figure 2.17(c) shows a grey scale representation of the linear conductance measured as a function of gate and perpendicular magnetic field B for the gate region of diamonds #5-10 in the stability diagram. The Coulomb peaks follow nearly parallel paths but small wriggles can be seen. The relative shifts are small since EC À ∆E and no shifts larger than ∼ ∆Eavg are expected as level crossings then occur. Systematic shifts can, however, be identified in panel (d) where peak the separations, ∆Vg , have been extracted. For low fields (B . 0.3 T) the addition energies alternately shrink and grow consistent with an alternating spin filling. The dashed line shows the expected shift magnitude for g = 2 (with the above value for the gate coupling coefficient α) in good agreement with the measurement. By comparing with Ref. [30], this value of the g-factor corresponds to a very small dot size15 which, as discussed below, is not consistent with a simple capacitance estimate based on the charging energy. This discrepancy may result from coupling between the levels causing a suppression of the shifts. Another explanation may be that the dot geometry is not expected to be a well-defined segment of the wire, as is the case in the dots of Ref. [30], which are confined by growth-defined barriers. For larger fields a complicated level crossing behavior is observed in Fig. 2.17(d) which will not be discussed further. Nanowire quantum dot geometry The results presented above were all obtained from measurements on one device which was particularly thoroughly investigated, however, measure15 In Ref. [30] g ≈ 2 was found for dots defined in a wire segment of length 8 nm in a wire of diameter 70 nm. 35 L ~700nm G [a.u.] 15 Cg [aF] 10 5.7 5.8 Vg [V] 5.9 L 5 0 0 250 500 750 L [nm] Figure 2.18: The capacitance Cg of the back gate to the quantum dot as a function of the electrode separation L. The linear relation shows that the dot size is determined by the electrode spacing rather than the random defects along the wire. The linear fit extrapolates to zero at a finite L indicating a substantial contact induced depletion of the wires. The rightmost inset shows an SEM micrograph of a large L ∼ 700 nm device (scale bar, 1 µm) and the leftmost inset shows its regular Coulomb peaks in the linear conduction G as a function of gate voltage at T = 300 mK. ments on other devices were found to exhibit similar characteristics. An important feature is the striking similarity in the dot characteristics in the measurements of Fig. 2.16 and Fig. 2.17 despite the fact that they were performed at significantly different gate voltages (separate ends of Fig. 2.15(a) corresponding to addition of ∼ 150 electrons). This shows that the dot confinement is relatively insensitive to the electrostatic environment and that the wire behaves as a single dot over the whole range. Such behavior is generally observed in these devices, however, in a few cases an additional period in the Coulomb peaks was found at the highest gate voltages characteristic of two dots in series. Since no special efforts were made to tailor the confinement of the dot, such as growth defined barriers or electrostatic gates, it is difficult to conclude about the geometry from the above measurements. Some information is, however, conveyed when using the CB peak spacings ∆Vg ∼ 40 mV to calculate the gate capacitance which, for ∆E ¿ EC , is given by (eq. 2.2) Cg ≈ e/∆Vg ∼ 4aF. (2.4) This can be repeated for all the devices and Fig. 2.18 shows Cg as a function of the electrode spacing L. The rightmost inset shows an SEM image of the 36 device having the largest electrode spacing L ∼ 700 nm and the leftmost inset shows G vs. Vg for this device at 300 mK, showing the regularity of the Coulomb peaks. As seen, the capacitance scales linearly with L for these devices, showing that the size of the dot is determined by the electrode spacing rather than random defects or potential fluctuations along the wire, which would result in random dot sizes. The linear fit extrapolates to zero at a non-zero electrode spacing ∼ 120 nm and we speculate that the dot is defined in a middle segment of the wire with barriers induced by depletion of the wire near the contacts (i.e., the 120 nm corresponds to the depleted segment). Assuming that the dot extends the entire cross section of the wire, we can compare the capacitance with a model of a cylinder with radius rnw ≈ 35 nm and length lnw situated h ≈ 535 nm (center to plane) above a conducting back-plane. For this geometry the capacitance is C0 = 2π²0 ²r lnw = 0.06 aF/nm × lnw , ln(2h/rnw ) (2.5) where ² = 3.9 is the dielectric constant of the silicon dioxide substrate. For Cg = 4aF this gives lnw ≈ 65 nm for the effective length of that segment of the wire which comprises the quantum dot. This is substantially shorter than the electrode separation L = 300 nm in qualitative agreement with the above picture. In Ref. [10] similar contact-defined barriers were found for quantum dots based on silicon nanowire devices and it is often the case for single-wall carbon nanotube devices[43]16 , but due to the unusual contact properties of InAs such regular behavior comes as a surprise. More work (adding data points to Fig. 2.18 from additional device batches) is needed to substantiate whether this behavior is indeed a general property of the MBE grown InAs nanowires. Tunable nanowire devices in the closed dot regime The results discussed above were all measured for the devices which were fabricated to have large barriers (type-2, cf. the introduction). The type-1 devices exhibit tunnel barriers which are tunable with the back-gate and below, the measurements in the intermediate coupling regime are presented. In the closed dot regime, these devices are unstable and exhibit multiple gate-switches. However, for completeness a stability diagram measured for a type-1 device, in the close-dot regime, is shown in Appendix C for a Vg range covering a few Coulomb diamonds. The behavior is qualitatively similar to the results presented in this section, however, the charging energies and level spacings are generally larger, EC ≈ 6 meV and 0.3 . ∆E . 1 meV, respectively. 16 In these cases the dot size fits the electrode separations, i.e., the depletion is negligible. 37 Summary: InAs nanowire devices in the closed dot regime In summary, we have, in this section, treated nanowire devices in the closed dot regime exhibiting quantum dot physics. Initially the constant interaction model was discussed providing the terminology for the presentation of the wire data. Subsequently, low temperature measurements (T = 300 mK) of a device with L = 300 nm electrode separation were treated in detail. The results show that the wire acts as a stable and regular quantum dot over the entire gate voltage range corresponding to the addition of ∼ 150 electrons. The characteristic energies were found to be EC ∼ 2.3 meV for the charging energy, and ∆Eavg ∼ 0.1 meV for the average level spacing. In one Vg -region an alternating pattern of addition energies were found, caused by the presence of two-fold (most likely spin) degenerate levels. In a different Vg -region the magnetic field dependence of the addition energies was investigated showing behavior consistent with an alternating spin for the ground states at zero field. In the final paragraph the geometry of the quantum dot was discussed. It was shown that the back gate capacitance Cg , being a measure of the dot size, scales linearly with electrode separation and the results indicated a substantial depletion of the wire at the contacts. Furthermore, by comparing with a cylinder-plane model for the capacitance, the data was found to be consistent with the dot being confined to a wire segment considerably shorter than the electrode separation. More work is needed to support this picture. The results in this section were all for devices in the regime of weak coupling to the electrodes where transport occurs by sequential tunneling events, i.e., an electron must tunnel off the dot before the next can tunnel on. In the upcoming section we treat devices in the intermediate-coupling regime where also higher-order processes contribute significantly, leading to additional characteristic features in the transport measurements. 38 2.5 Kondo physics in InAs nanowires This section treats measurements on InAs nanowire devices where the coupling to the leads is stronger than was the case for the results treated in the previous section. The barriers are still sufficiently opaque that Coulomb blockade plays a substantial role, however, higher order processes contribute significantly to the transport in this regime, leading to the observation of features associated with cotunneling and the Kondo effect. The present results are the first to demonstrate the presence of the Kondo effect in nanowire systems and the main contents of this section has been presented in Ref. [44]. Before discussing the measurements, however, we first consider the physical origin of these phenomena. Co-tunneling In the closed dot regime of the previous section, the number of electrons residing on the dot is a well-defined integer number N , and at zero bias transport only occurs at discrete degeneracy points. Measuring G vs. Vg results in narrow peaks separated by stretches of zero conductance corresponding to Coulomb blockade. A schematic of the blockaded situation is repeated in Fig. 2.19(a). Classically, transport is blocked as it requires a finite energy for the electron on the dot to tunnel off (or for an additional electron to tunnel on). Quantum mechanically, however, such an event, leading to the intermediate virtual state depicted in panel (b), can occur as long as the system only exists in the virtual state for a time ∆t sufficiently short not to violate the Heisenberg uncertainty relation ∆t . ~/EC (the charging energy EC being the energy associated with the charge fluctuations). Thus, as long as another electron tunnels onto the dot again within ∆t, thereby returning the dot to its original state, a current can flow by the cotunneling sequence (a)-(c), leading to a non-zero conductance in the Coulomb valleys (diamonds) for stronger coupling[45]. An example of such behavior in a nanowire device was already given in the introduction (Fig. 2.4). Since the above sequence occurs at zero bias and leaves the dot in the ground state the process is called elastic cotunneling. For larger bias voltages, inelastic cotunneling processes, which leave the dot in an excited, state become accessible as illustrated in the schematic in panel (d). The onset occurs at eVsd = ∆E (the excitation energy of the dot) and in a bias spectroscopy measurement, such processes result in features horizontally truncating the Coulomb diamonds as illustrated in panel (e). The Kondo effect The transport properties of bulk alloys containing magnetic impurities (e.g., Fe in Au) have been studied for almost a century. In the 1930’ies it was found 39 (a) (b) Virtual .. . .. (c) .. . .. (d) .. . .. (e) N-1 N N+1 Vsd Vsd .. . .. Vg Figure 2.19: Schematic illustration of cotunnel processes. In the sequence (a)-(c) an electron is transferred from source to drain electrode through an intermediate virtual (classically forbidden) state. The process is elastic leaving the dot in the ground state. (d) Example of an inelastic cotunneling process leaving the dot in an excited state. Such processes become accessible when the bias provides the required energy Vsd = ∆E. The consequence for the stability diagram is schematically illustrated in (e). In the middle of the diamond only the elastic processes are possible (light gray) and above Vsd = ∆E also inelastic cotunneling contribute (dark gray). that the temperature dependence of the electric resistivity was described by the relation ρ(T ) = ρ0 + aT 5 − b log(T ). The first two terms were well understood to arise from scattering off static impurities (T -independent) and of phonons (which ”freeze” out at low temperatures), respectively. A satisfactory explanation for the logarithmic term, which causes an increase in the resistivity at low temperature, was not given until the 1960’ies. At that point Jun Kondo showed that the measurements could be explained by considering multiple spin-flip scatting of the conduction electrons on the impurity spin[46]. In this way the conduction electrons screen the impurity spin leading to the formation of a many-body Kondo state (Kondo cloud) accompanied by a resonance at the Fermi level. This increased scattering results in the observed resistivity increase below a temperature TK being the characteristic binding energy of the Kondo state17 . This original observation of the Kondo effect is concerned with bulk alloys containing many magnetic impurities. Consider now a quantum dot hosting an odd number of electrons. In the simplest case, pairs of electrons with anti-aligned spins are formed, leaving a single unpaired spin-1/2 in the dot, 17 For further material on the Kondo effect in bulk metals see Refs. [47, 48, 25] 40 (b) (a) (d) Virtual (c) (e) DOS TK N Even Vsd N+2 N+1 Vg Γ Figure 2.20: Schematic illustration of the processes leading to the Kondo effect in odd-N quantum dots. In (a)-(c) an electron is transported through the dot while flipping the dot spin. The sum of all such processes lead to a many-body Kondo state appearing as a sharp resonance in the density of states at the Fermi level leading to an enhanced conductance at zero bias. At higher biases this enhancement is lost and the Kondo appears in a stability diagram measurement as illustrated in (e). as illustrated in Fig. 2.20(a). In this case the dot and the metallic electrodes constitute a system in many ways analogous to a single magnetic impurity embedded in a metallic host, and indeed the two systems share the intriguing phenomenon of the Kondo effect. As for the bulk alloys, the dot spin is screened by the conduction electrons through a large number of spin-flip cotunnel events. An example of such a process is schematically illustrated by the sequence (a)-(c) of Fig. 2.20. This again leads to a correlated many-body state extending into the leads, however, contrary to the bulk case, the increased scattering contributes to the current and thus leads to an enhanced conductance. Thus, the Kondo effect will appear as a conductance increase in the odd-N Coulomb valleys upon lowering the temperature. This is opposite to the behavior expected for ordinary Coulomb blockade which dominates in the even-N valleys where the net spin (in the simplest case) is zero and no Kondo effect is expected. The Kondo state leads to a peak of width ∼ TK in the density of states locked to the chemical potentials of the leads as illustrated in Fig. 2.20(d). In the non-equilibrium situation at finite bias, the Kondo enhanced conductance is lost and in a stability diagram measurement, the signature of the Kondo effect is therefore a zero-bias conductance enhancement through the odd-N diamonds as 41 illustrated in panel (e). Another distinct feature of the Kondo effect is the splitting of the Kondo-ridge in a magnetic field B into two peaks at finite bias eVsd = ±gµB B where g is the g-factor and µB = 58µ eV/T, the Bohr magneton[49, 50]. Investigating the Kondo effect in a quantum dot allows much extended flexibility compared to bulk systems. First of all, such measurements probe a single impurity spin rather than an ensemble as in the bulk diluted alloys. Furthermore, it allows individual tuning of various parameters such as the coupling to the leads Γ, the distance from the Fermi level to the impurity level ²0 (controlled by the gate), and the possibility of investigating non-equilibrium effects by applying a finite bias across the impurity. For example, midway along the Kondo ridge (Fig. 2.20(e)) the Kondo temperature is given in terms of these parameters as[47] 1 1 kB Tk = (ΓEC ) 2 e−πEC /2Γ (2.6) 2 For these reasons, the Kondo effect has received renewed attention since it was first discovered in quantum dots in 1998 by Goldhaber-Gordon et. al. [51]. This pioneering work was based on dots defined by electrostatic gating of the 2DEG in a GaAs/GaAlAs heterostructure. Subsequently the effect has been observed in devices based on single and multi-wall carbon nanotubes[52, 53], single molecules[54, 55], fullerene peapods[56], and the present chapter adds the InAs nanowire system to this list[44]. For further details on the Kondo effect in quantum dots see Refs. [25, 47, 49, 57, 58]. Kondo physics in InAs Nanowire quantum dots This section considers the experimental evidence for the existence of the Kondo effect in the InAs nanowire quantum dots. The Kondo effect is only visible for temperatures lower than TK and as seen from Eq. 2.6 making TK large requires a strong coupling to the leads. The results considered in the previous section were measured on devices which had large contact barriers due to the fabrication scheme (type-2). In this section we consider the (type1) devices where the wire surfaces received both oxygen plasma etching and a HF wet-etch (immediately) prior to evaporation of the contacts (cf. section 2.3). It turns out that the contact transparency in these devices varies with gate potential rendering it possible to investigate the intermediate coupling regime relevant for the Kondo effect. Figure 2.21 shows the temperature dependence of the G vs. Vg trace for a typical type-1 device. At high temperatures, the conductance varies monotonically from 0.9 e2 /h at Vg = −10 V to 2.8 e2 /h at Vg = 10 V thereby identifying, as for the type-2 devices, the carriers as n-type. At lower temperatures, the slope of the G(Vg )-trace increases but the conductance at Vg = 10 V remains effectively unchanged. This is in contrast to the behavior of the type-2 devices (Fig. 2.10) and indicates a low barrier at the 42 5 0 G[e2/h] -8.9 2 -8.8 -8.7 V [V] g G [e2/h*10-2] 3 -8.6 270K 130K 15K 0.3K 1 0 -10 -5 0 Gate [V] 5 10 Figure 2.21: The linear conduction G vs. Vg for temperatures T = 0.3 K, 15 K, 130 K, 270 K for a type-1 device showing gate-dependent barriers: At 0.3 K and Vg > 2 V the contacts are open and reproducible conductance fluctuations are observed. For −6 V < Vg < −1 V the Kondo effect is observed and for Vg < −7 V the device is dominated by CB as expanded in the inset (see text). contacts at Vg = 10 V increasing with smaller gate-voltages. At 0.3 K, the behavior is as follows: For Vg . −7.5 V (inset) the device behaves as a quantum dot in the CB regime with large tunnel barriers between the leads and the dot with transport exhibiting sharp peaks separated by regions of zero conductance as discussed in the previous section. For Vg & 2 V large-scale reproducible conductance oscillations are observed. There are no indications of charge quantization (Coulomb blockade), supporting the view of low barriers between the leads and the wire in this gate range. This regime will be treated in the upcoming section, where the fluctuations are interpreted by electron interference effects. The focus of this section is on the intermediate gate region where broad CB peaks appear (see also Fig. 2.4). The qualitative behavior of Fig. 2.21 is typical for the type-1 devices, however, the gate-voltage scale may change between devices and between subsequent cool-downs of the same device due to sensitivity to the charge configuration in the substrate. Also variations are found in the size of the gate interval in which the device is in the intermediate coupling and closed regimes, respectively. In Fig. 2.22(a) the G vs. Vg trace for a short gate range has been enhanced. Solid and dashed traces show measurements for T = 300 mK and T ≈ 800 mK, respectively. Broadened CB peaks are observed and as in the previous section each valley corresponds to a fixed number of electrons (N ) on the dot. The non-zero valley conductance Gv indicates a significant contribution to the conductance from elastic co-tunneling processes and Gv exhibits an alternating pattern where high-Gv valleys are followed by valleys 43 G [e2/h] 1.0 75 0 Valley # 4 8 (c) ∆Vg [mV] 100 ∆Vg [mV] (a) 0.5 #0 100 90 80 70 0.0 -4.50 -4.25 0 -4.00 0.25 B [T] Vg [V] 0.5 (b) B [T] 1.5 1.0 0.5 0.0 -4.50 -4.25 Vg [V] -4.00 (d) 2.0 Vsd [mV] E O E O E O E O 1.0 0.0 -1.0 (c) -2.0 -4.50 -4.25 Vg [V] -4.00 Figure 2.22: (a) Measurement of the linear conduction as a function of gate voltage in the intermediate regime in Fig. 2.21 for T = 300 mK (solid) and T = 800 mK (dashed) showing 8 broad Coulomb peaks. The arrows indicate the temperature dependence of the valley conductance for increasing T . Inset shows the peak separation identifying a clear even-odd pattern due to spin degeneracy. (b) The linear conductance at T = 300 mK as a function of gate voltage and perpendicular magnetic field B showing the evolution of the CB peaks (darker = more conductive). In (c) the low-field peak separations have been extracted establishing the odd-N electron occupation for the narrow CB valleys. (d) Corresponding stability diagram at T = 300 mK showing 8 CB diamonds numbered E/O corresponding to even/odd N , respectively (darker = more conductive). In each odd-N diamond a high conductance ridge is observed at zero bias (black arrow) due to the Kondo effect. The white arrows point to examples of features attributed to inelastic cotunneling. 44 of lower Gv and vice versa. This even-odd pattern repeats in the peak separations ∆Vg shown in the inset extending 9 Coulomb diamonds. As for the even-odd effect discussed in connection with Fig. 2.16, the corresponding even-odd pattern of the addition energies suggests a twofold degeneracy of the dot levels. The electron spin is identified as the origin of this degeneracy by measuring the evolution of the peaks in a magnetic field. Figure 2.22(b) shows the linear conductance measured as a function of Vg and perpendicular magnetic field B analogous to the measurement for closed dots shown in Fig. 2.17(c). Valleys corresponding to odd(even) N are expected to widen(shrink) by the Zeeman splitting gµB B[59] and at low fields this pairing behavior is readily observed in panel (c) where the peak separations have been extracted from (b)18 . The complicated pattern observed at higher fields in (b) is due to level-crossing/anti-crossing and will not be discussed further. Thus, in the valleys in Fig. 2.22(a) corresponding to the short ∆Vg , the dot holds an odd number of electrons N , and the Kondo effect can be anticipated. As seen, these valleys indeed show an increased conductance with respect to their neighboring valleys. By considering the T = 800 mK trace (dashed line) the conductance in the even-N valleys is seen to increase upon heating the sample (arrows) as expected for normal Coulomb blockade. Contrary, in the odd-N valleys Gv exhibits the opposite behavior showing that transport occurs by different mechanisms in the even and odd-N valleys. These observations agree with the presence of the Kondo effect and in the stability diagram for this region, shown in Fig. 2.22(d), there are indeed pronounced high-conductance ridges appearing at zero bias in every odd-N diamond (one is indicated by a black arrow). Temperature dependence of the Kondo ridge The temperature dependence of a Kondo peak is described by the interpolation function ³ T 02 ´s Gv (T ) = G0 2 K 02 (2.7) T + TK 0 = T /(21/s −1)1/2 and s = 0.22 expected for a spin-half system[61, Here TK K 49]. The temperature dependence thus allows for an estimate of the Kondo temperature. As seen in Fig. 2.23(a), however, only a small increase in valley conductance is observed upon lowering the temperature from 800 mK to 300 mK and it has not been possible to make a meaningful fit to the above formula for these data. A crude estimate of the coupling Γ ≈ 0.7 meV from 18 We note, that an estimate of the g factor from the peak separations is not straight forward since the Kondo effect (which will be seen below to be present for the odd-N diamonds) modifies the peak separations[60] and as a magnetic field affects the Kondo ridge, it contributes to the behavior in Fig. 2.22(c). Below it will be shown that the splitting of the Kondo ridge in a magnetic field allows for determination of g. 45 T [K] (a) 1.5 0.3 3.0 (b) 2 1.0 Vsd [mV] dI/dVsd [e2/h] 0 0.5 x1/2 0.0 -2 ΓK -2 2 0.3K 0 -2 2 0.9K 0 1.6K -2 0 Vsd [mV] 2 -3.9 -3.7 Vg [V] -3.5 Figure 2.23: (a) Top curve: temperature dependence of the conductance for the strong Kondo ridge. The solid line is a fit to the formula given in the text with a Kondo temperature of 2.1 K. Lower curve shows dI/dVsd vs. Vsd through the middle of the ridge (scaled by a factor 0.5). The peak width corresponds to TK = 2.2 K. (b) Stability diagrams measured for T = 0.3 K, 0.9 K, 1.6 K (top, middle, bottom) showing a particular strong Kondo ridge at 300 mK (intersected a by dashed line). At higher temperatures the Kondo ridge disappears but the Coulomb diamond pattern prevails. the width of the Coulomb peaks gives an estimate of TK ≈ 0.8 K from Eq. 2.6. The top panel of Fig. 2.23(b) shows a stability diagram measured for T = 300 mK in a different cool-down of the same device. A particularly strong Kondo resonance is observed, and Fig. 2.23(a) shows dI/dVsd vs. Vsd through the middle of the ridge at 300 mK (lower curve) showing a pronounced Kondo peak at zero bias. The temperature dependence of the peak conductance is shown in the upper curve (circles) and the solid line shows a fit to equation 2.7. The agreement is excellent and the fit yields TK = 2.1 K and s = 0.22 supporting the Kondo nature of the ridge. The Kondo temperature can also be estimated from the full width at half maximum ΓK ≈ 2kB Tk /e of the Kondo peak and for the data in Fig. 2.23(a) we find ΓK = 340 mV and thus TK = ΓK e/2kB = 2.2 K in good agreement with the estimate above. The lower and middle panels of Fig. 2.23(b) show the stability diagram measured at higher temperatures T = 900 mK and T = 1.6 K, respectively. The Kondo ridge diminishes with increasing T and at 1.6 K only the Coulomb diamond pattern persists. In the case of equal coupling to the leads the conductance in the Kondo valley is expected to saturate for T ¿ TK at the ideal value G0 = 2e2 /h (unitary 46 (b) (a) 0.2 0.5T -2.2 -2.5 G [e2/h] -3 0T -2.5 Vg [V] -2.2 Vg [V] 0.1 0T 0.0 δΚ gFIT 1.5 0 ∆V [mV] Vsd [mV] 3 ∆0 (mV) ∆ Κ (g ) 7.9 +- 0.1 0.2 +- 0.01 δΚ (c) 7.7 +- 0.1 -0.02 +- 0.01 ∆c (o) 8.5 +- 0.7 -0.03 +- 0.03 1.0 Vsd=Vc 0.5 0.9T ∆c Vsd=0 0.0 0.0 ∆Κ -1 0 1 0.5 1.0 1.5 B [T] Vsd [mV] Figure 2.24: (a) Stability diagrams at B = 0 T (leftmost inset) and B = 0.5 T (rightmost inset) of an odd-N diamond with a Kondo resonance. Main panel shows traces through the middle of the diamond in perpendicular magnetic fields of 0 T, 0.1 T, . . . , 0.9 T (each off-set by −0.0075 e2 /h). The peak at Vsd ≈ 1.25 mV indicates the on-set of inelastic cotunneling. (b) The separation of the peaks in magnetic fields as indicated in (a). The upper inset shows the g-factor and extrapolated splitting ∆0 at B = 0 T for the linear fits. Rightmost inset: Schematic diagram of cotunneling processes for zero source-drain bias (left) giving the usual Kondo effect, and at finite bias (right) yielding Kondo enhanced cotunneling peaks out of equilibrium. All states have total spin S = 1/2. All results are for T = 300 mK. limit). We have not for any of the investigated Kondo ridges observed a saturation of Gv within the accessible temperature range T & 300 mK. For the Kondo ridge in Fig. 2.23, Gv seems to approach a value . 1.5e2 /h which is smaller than the ideal value probably due to asymmetric coupling to the leads in which case G0 is modified by a factor ∝ ΓL ΓR /(ΓL + ΓR )2 . Magnetic field dependence of the Kondo ridge In a magnetic field the spin-degenerate states of the quantum dot are split by the Zeeman splitting gµB B, which for Kondo resonances leads to peaks at Vsd = ±gµB B/e, independent of gate voltage[49]. The stability diagrams inset to Fig. 2.24(a), show a Kondo ridge through an odd-N diamond measured at zero field (leftmost inset) and at B = 0.5 T (right). Indeed, a gate independent splitting of the ridge is clearly observed. The main panel shows dI/dVsd vs. Vsd for Vg corresponding to the middle of the diamond and for different magnetic fields B = 0−0.9 T. As expected the peak splitting is proportional to B. Different methods for determining g from the splitting of the Kondo peak have been suggested in the literature. Early works used the distance δK between the peaks in dI/dV [50, 60], however, subsequent theoretical work has suggested the use of the distance ∆K between peaks in 47 d2 I/dV 2 (steepest points in dI/dVsd )[62] or a combination of the two[63]. Figure 2.24(b) shows ∆K (open squares) and δK (solid squares) extracted from the data of panel (a) and measurements at higher fields (not shown). Both ∆K and δK show a clear linear dependence and the parameters of the linear fits are shown in the inset with g calculated from the slope and ∆0 being the extrapolated splitting at B = 0 T. Both methods gives g ≈ 8 considerably smaller than the g-factor of bulk InAs (gbulk = 15) but larger than the value found in the previous section. The value is, however, as discussed earlier expected to be strongly dependent on dimensionality and size of the system[64, 30]. As discussed above, inelastic cotunneling processes, which leave the dot in an excited state, set in when Vsd reaches the excitation energy of the dot. This gives rise to horizontal ridges at finite bias within the Coulomb diamonds as seen in Fig. 2.22(d) and Fig. 2.24(a). At zero field, Fig. 2.24(a) shows a sharp peak appearing at this onset (Vc ≈ 1.25 mV), in addition to the zero bias Kondo peak (odd N ). In a magnetic field, the doublet excited state splits by ∆c = gµB B, i.e., half that of the Kondo peak, allowing for an independent estimate of g. The splitting is readily observed, and the linear dependence of ∆c vs. B extracted in panel (b) (circles) yields a g-factor of 8.5 ± 0.7 consistent with that measured from the Kondo peak. Figure 2.25 shows the splitting of the 6 Kondo ridges from two samples that have been investigated in a magnetic field. For all the ridges the splitting is close to that of Fig. 2.24 (also included in Fig. 2.25). The small variation in g is interpreted as a result of slightly different dot geometries/sizes. Finally it is noted that non-equilibrium population of the excited states at finite bias[62] may cause a broad cusp at the cotunneling onset. However, the peak in Fig. 2.24(a), being considerably narrower than the threshold bias, is inconsistent with this mechanism. We speculate that the finite bias peak is a signature of a Kondo resonance existing out of equilibrium. For carbon nanotubes, a detailed quantitative analysis recently showed that for an even-N quantum dot, inelastic cotunneling processes can result in a nonequilibrium singlet-triplet Kondo effect, accompanying transitions between the singlet ground state (S = 0) and an excited triplet state (S = 1)[65]. The Kondo correlations are in this case indicated by peaks at the cotunneling onset being narrower than the threshold bias. We see similar, sharp peaks in our devices in even-N diamonds, however, for the odd-N example in Fig. 2.24(a), the peak is intensified by (spin-flipping) transitions promoting the dot into excited states with the same total spin S = 1/2 as the ground state19 . More experimental and theoretical work is needed to substantiate these findings. 19 For a dot with nearly equidistantly spaced levels such as the region analyzed in Fig. 2.22(a) (insert), two different excited orbital states can contribute to this resonance, cf. the schematic in Fig. 2.24(b) 48 2.0 0T 0.3 G [e2/h] 0.0 1.5 Vg [V] 1.0 1.1 2 1.0 -0.5 0 Vsd [mV] 0.5 0 Vsd [mV] ∆V [mV] -0.3 1T 0.5 0T -2 2 0 -2 0.0 0.0 0.5 1.0 B [T] 1.5 0.25T 2.0 Figure 2.25: The splitting of 6 Kondo ridges in a perpendicular magnetic field. Rightmost insets show a stability diagram measured with B = 0 T (top) and B = 0.25 T (bottom). A zero bias Kondo ridge is seen in the top inset, and a gate-independent splitting is seen in the bottom inset. Leftmost inset shows dI/dVsd vs. Vsd through the middle of the diamond in the stability diagrams for B = 0 − 1 T (curves offset by −0.0075(e2 /h) ∗ (B/T)). A linear splitting is clearly observed as shown in the main panel with leftpointing triangles (C) along with the splittings of all the Kondo ridges that were investigated in a magnetic field. The data of Fig. 2.24 is shown with squares. The points fall almost along the same line showing that the g-factor g ∼ 8 only varies slightly in these samples. Summary: Kondo effect in InAs-nanowire devices In summary we have in this section treated nanowire devices in the intermediate coupling regime. Evidence of the higher order transport processes of inelastic cotunneling and the Kondo effect was provided, and the latter was treated in detail. From the splitting of the Kondo ridge in a perpendicular magnetic field a g-factor of ∼ 8 was inferred for all the Kondo ridges that were studied in a magnetic field (6 ridges from 2 devices). 49 2.6 Conductance Fluctuations in InAs nanowires In the previous sections the phenomena associated with nanowire devices with low and intermediate coupling to the leads were investigated. Continuing along this line we treat, in this section, the regime of vanishing barriers at the contacts. At low temperature the device in this regime acts as a diffusive (or quasi-ballistic), phase coherent conductor. Due to electron interference the conductance fluctuates as a function of gate potential and magnetic field as these change the wavelength and phase of the electrons. Below, we first investigate the fluctuation as a function of Vg and then study the effect of a magnetic field on the average conductance, where an average over impurity configurations is achieved by averaging over Vg . In this way it has been possible to study the phenomena of weak (anti) localization in single wires. Conductance fluctuations In the discussion of Fig. 2.21 it was already mentioned that the open regime is relevant for type-1 devices at the highest gate-voltages. Figure 2.26(a) shows the linear conductance as a function of gate voltage and temperature for such a device with L ≈ 300 nm electrode separation measured on the SEM micrograph in the inset to panel (b)20 . The conductance is high ∼ 2e2 /h and at low temperatures there is no trace of Coulomb blockade oscillations. The conductance exhibits large-scale fluctuations as a function of gate which are reproducible in the sense that the exact same trace is found upon repeating the measurement21 . Indeed, the corresponding stability diagram shown in panel (b) shows no Coulomb blockade diamonds but instead an irregular pattern of low-conductance regions appear. The irregularities contrasts the regular Fabry-Perot interference pattern found in ballistic nanotube wave guides[66] where reflection only occur at the contacts, but our results bear great resemblance to results found for high conductance diffusive or quasi ballistic nanotube samples[53, 67]. The results of Fig. 2.26 are interpreted as an effect of quantum interference due to scattering off random impurities along the wire - the pattern arises since, changing the Fermi wavelength by means of Vg , has the same effect as reconfiguring the impurity distribution and thus changes the transmission probability[42] (similar changes are expected upon applying a magnetic field). Considering a conductor of length L and mean free path `e as a sum of L/`e uncorrelated p fluctuating segments the root-mean-square fluctuations δGrms = Var(G) p of the conductance is a factor `e /L smaller than the average conductance hGi and thus for L À `e no significant fluctuations would be expected. 20 The measurements of the Kondo effect in Fig. 2.23 are from this device at a lower gate-voltage. 21 Except for occasional hysteretic effects 50 (c) 273K 68K 35K 2 18K 1 11K 7K 4K 0. 3K 0 (a) 2 4 6 Gate [V] 8 10 δGrms [e2/h] G [e2/h] 3 0.1 1µm 0.01 0.1 1 T [K] 2 0 0.3 dI/dVsd [e2/h] 2.7 (b) Vsd [mV] 10 -2 2 4 3 5 Gate [V] Figure 2.26: (a) The linear conductance as a function of gate voltage for different temperatures for a type-1 nanowire device with vanishing contact barriers and L = 300 nm electrode separation. Upon lowering the temperature large reproducible conductance fluctuations develops due to quantum interference. (b) Differential conductance measured at T = 300 mK as a function of gate and source-drain bias. An irregular pattern of low and high G regions is observed. (c) The root-mean-square fluctuations δGrms of G(Vg ) as a function of temperature after subtracting a second order polynomial (see text). δGrms saturates at low temperature to ∼ 0.5 e2 /h - dashed line is a guide to the eye. Inset to (c) shows an SEM micrograph of the device. Quantum correlations, however, extends over the phase coherence length `ϕ which at low temperatures can be much longer than `e and for `ϕ À L and zero temperature, it is a remarkable result that fluctuations on the order of e2 /h are expected regardless of the degree of disorder or sample size. For this reason these fluctuations are often called universal conductance fluctuation (UCF) (see Ref. [42] and references herein). For `ϕ < L the sample can be considered as consisting of L/`ϕ uncorrelated segments and the fluctuations are suppressed by a factor (`φ /L)3/2 [42]. As seen in Fig. 2.26 the G(Vg ) trace is monotonic at higher temperatures with a slightly positive slope due to the changing carrier concentration and the fluctuations develop upon lowering the temperature (and increasing `φ ). Panel (c) shows δGrms calculated from the traces in (a) after subtracting a second order polynomial to account for the monotonic background. Indeed δGrms increases rapidly for decreasing T and saturates around T . 1 K at a 51 value ∼ 0.5 e2 /h relatively close to the expected value ∼ 0.73 e2 /h for a conductor which is 1D with respect to `φ and wire length much larger than the diameter[42]. Thus the results indicate that `φ & L = 300 nm for T . 1 K and the appearance of the conductance fluctuations shows that the wires are diffusive (or quasi-ballistic) conductors (`e < L). Magneto conductance In this section we briefly consider the effect of applying a magnetic field on the transport in nanowire devices in the open regime. In view of the above considerations, random conductance fluctuations are again expected and, in addition, another effect of quantum interference, known as weak localization (WL), may be anticipated. The weak localization-correction to the conductivity is caused by the constructive interference of coherently backscattered time-reversed paths (leading to an enhanced probability for the electron to stay where it is). This enhanced backscattering probability reduces the conductance. Upon applying a magnetic field, the electrons along such two paths pick up different phases and the constructive interference is lost, leading to a positive magnetoconductance. In the presence of strong spin-orbit scattering the backscattering interference can be destructive leading to the opposite behavior in a magnetic field, an effect known as weak anti-localization (WAL)[68]. The low temperature magnetoconductance of InAs nanowires were studied in details by Hansen et. al. in Ref. [36]. A crossover from WL to WAL was found upon increasing Vg due to a decreasing spin relaxation length `so (relative to `φ ). The presence of the random conductance fluctuation impedes the analysis of WL and WAL in the magnetoconductance traces, and in Ref. [36] this obstacle was overcome by measuring 40 identical nanowire devices in parallel thus averaging out the UCF and leaving only the weak localization correction. Figure 2.27(a) shows the conductance as a function of Vg for an open nanowire device measured at 4.2 K showing conductance fluctuations. Panel (b) shows G as a function of perpendicular magnetic field −0.5 T ≤ B ≤ 1.5 T for four different gate voltages. As expected, G(B) is symmetric in B and exhibits large scale reproducible conductance fluctuations. A peak is observed at zero field for all Vg , but the peak width and height is modified by the conductance fluctuations. Instead of averaging out the fluctuations by measuring multiple devices in parallel, G was measured as a function of B and Vg 22 and subsequently averaged over Vg ∈ [−10, 10 V] yielding hGiVg as illustrated in panel (c). Note, that the measurement was only performed for one direction of B, but from panel (b), G is known to be symmetric in B, and in panel (c) the measurements have been repeated for both directions for 22 Measurement performed with Vg on the inner loop. 52 Vg = 10V (b) (a) 4.2 K -5V 2.0 -10V 1.0 4.2 K B = 0T -5 0 Vg [V] 5 10 <G> [e2/h] G [e2/h] 0V -10 (c) 1.65 3.0 1.60 4.2 K 0.0 0.5 1.0 B [T] 1.5 (-)0.5 0.0 B [T] 0.5 Figure 2.27: (a) Measurement of G vs. Vg at T = 4.2 K showing large reproducible fluctuations due to quantum interference. (b) Measurement of G as a function of perpendicular magnetic field B at four different gate voltages (marked in panel (a)). The conductance is symmetric in B and exhibits fluctuations due to quantum interference. (c) The average conductance hGiVg ∈[−10,10V ] as a function of B. Since G is symmetric (as confirmed in (b)) only the part B > 0 T is measured and the data is repeated for B < 0 T to show clearer the features attributed to WAL (negative magnetoconductance at low fields) and WL (positive magnetoconductance at higher B). The solid line is a fit to the theory (see text). clarity. A clear negative magnetoconductance is observed at low fields interpreted as an effect of weak anti-localization, thus showing that a significant spin-scattering takes place in these wires. At higher fields the interfering paths are shorter and the spin does not scatter significantly and the positive magnetoconductance of (normal) weak localization is recovered. Since the conductance changes significantly over the averaging interval, showing that the carrier density and possibly also the scattering lengths changes, a quantitative analysis of these data should not be taken too seriously. However, to get a rough estimate for the parameters which would result in a WAL curve like the one in Fig. 2.27(c) we fit the data to the model which was found in Ref. [36] to best describe the InAs nanowire results23 . In this model the magnetoconductance with respect to the value at zero field is given by 4 1 ´− 12 1 ³ 1 1 ´− 12 2e2 n 3 ³ 1 + + − + hL 2 `2ϕ 3`2so DτB 2 `2ϕ DτB 3³ 1 1 4 1 ´− 12 1 ³ 1 1 1 ´− 21 o + , + + + + + 2 `2ϕ `2e 3`2so DτB 2 `2ϕ `2e DτB ∆G(B) = − − 23 Equation 3 of Ref. [36]: ’pure’ model assuming `e greater than wire diameter and `φ ¿ L which is expected to hold for the present data because of the relative high temperature T = 4.2 K, i.e., above the saturation temperature of Fig. 2.26(c) 53 where D is the diffusion constant and DτB = C1 `4m `e /W 3 + C2 `2mp `2e /W 2 , with W being the wire diameter and `m the magnetic length `m = ~/eB. This yields • `φ = 220 ± 20 nm, • `so = 195 ± 20 nm, and • `e = 100 ± 15 nm, The fit is shown on the figure with a solid line. If we assume that the resistance of the device in the open regime is dominated by the resistance of the wire, rather than contact resistances24 , the change in G observed in Fig. 2.27(a) is attributed to a changing carrier density. Relating the elastic mean free path to the conductivity in a simple Drude model we find that the conductance-range 1 − 3e2 /h in Fig. 2.27(a) corresponds to a carrier density of ∼ 1016 − 1017 cm−3 . These values are close to what was found in Ref. [36] for CBE grown InAs nanowires25 but again we stress that these values should only be taken as a rough guide and more experimental work is needed to make a detailed comparison based on the magnetoconductance. The measurement show, however, the importance of spin-scattering and the possibility of eliminating the conductance fluctuations by averaging over Vg thus enabling measurements of weak localization on individual wires. The averaging interval should, however, be chosen in a region where G does not change significantly (e.g. as in Fig. 2.26(a))26 . 24 Without a four-terminal measurement it is not possible to directly evaluate the contact resistances and therefore we do not know if this is actually the case. 25 `φ ∼ 250 nm, `so ∼ 150 nm, `e = 60 nm - for a gate corresponding to a conductance per wire similar to the present case. 26 We note, that repeating the measurement of Hansen et. al. with many wires measured in parallel, giving the scattering lengths as a function of Vg , is not applicable for our devices since this requires good reproducibility in device performance and at present, a considerable difference is observed between devices (threshold etc.). Further control of the fabrication may enable such measurements. 54 2.7 Nanowire devices with superconducting leads In the previous sections we have investigated electrical transport in nanowire devices contacted by normal metals (N). As discussed in Section 2.3 one of the advantages of new mesoscopic systems, such as nanotubes and nanowires, is the wider range of possible contact materials as opposed to traditional mesoscopic devices based on a two dimensional electron gas in a semiconductor heterostructure. For instance, applying ferromagnetic contacts may enable studies of spin polarized transport in mesoscopic devices[69, 70] and other exciting phenomena are anticipated if, instead of the normal metals, the wires/tubes are contacted by superconductors (S). In this case the nanowire provides a tunable link between the two superconductors and the system naturally lends itself for studies of the effects of size quantization, charge quantization, and electron interactions in transport between superconductors[71, 72, 73, 74, 75, 76, 77, 78, 21, 22, 23]. This section presents results on S-Nanowire-S junctions in the three regimes that were treated in the past sections for the case of normal metal leads (Coulomb Blockade, Kondo, UCF). Due to time limitations while performing the measurements the main body of results have been obtained in the Kondo regime, allowing only for a brief discussion of the closed-dot and open regimes. Future experiments are planned in order to investigate these regimes further. SNS junctions Superconductor-Normal metal-Superconductor (SNS) junctions (weak links) have been studied extensively in the past decades[79] and the most wellknown property of such junctions is probably that, in the absence of an applied bias, a dissipationless supercurrent Is = Ic sin(∆ϕ) can flow through the junction; an effect known as the dc Josephson effect[80, 81]. Here ∆ϕ is the phase difference between the two superconductors and Ic is the critical current the junction can sustain. Applying a voltage V across the junction results in a high frequency alternating current since the phase difference evolves according to the ac Josephson relation d(∆ϕ)/dt = 2eV /~. Thus, a current biased setup is required in order to measure the supercurrent in a weak link, and this has recently been achieved for both nanowire and nanotube devices[77, 78, 22, 21]. The supercurrent has not been the focus of the studies presented in this section and the measurements have been carried out in a voltage biased setup like those of the preceding sections enabling the investigation of another interesting feature of superconducting weak links, the subharmonic gap structure (SGS). The SGS is observed at finite bias and is caused by the process of Andreev reflections. An electron in the normal region impinging on the superconductor interface with an energy below the superconducting gap, cannot directly enter the superconductor since there are no available states in the gap. It 55 (a) ∆ (b) 2∆/3 Figure 2.28: Schematic illustration of the process of Andreev reflections (AR) in an SNS junction. For each superconductor the density of states is shown and solid/open circles represents electrons and holes, respectively. (a) Illustration of the process which opens at Vsd /e = ∆ involving a single AR. (b) Process involving two subsequent AR’s becoming possible at Vsd /e = 2∆/3. Only processes starting with an electron are shown. can, however, be transmitted as a Cooper pair if, at the same time, a hole is reflected. This process is known as an Andreev reflection (AR) and due to energy conservation, the reflection occurs symmetrically around the Fermi level as schematically illustrated in Fig. 2.28(a). In an SNS junction the reflected hole can subsequently be Andreev reflected as an electron at the other NS-interface as illustrated in Fig. 2.28(b). Such processes are known as multiple Andreev reflections (MAR) and from simple schematics as those in Fig. 2.28 it can be realized that upon increasing the bias of the junction, MAR processes involving n − 1 AR’s becomes accessible at bias voltages Vsd = 2∆/ne, n = 1, 2, ... while at the same threshold MAR’s processes involving n+1 AR’s are closed and therefore the IV-characteristics (or dI/dV ) will contain structure at these voltages - the SGS[80]. As we will discuss below, further considerations are needed if the normal region of the junction constitute a quantum dot with conduction through discrete levels and possible interactions of the AR with the electrons on the dot[73, 76, 77]. Sample preparation The samples for these studies were prepared in very much the same way as described in section 2.3 for the devices with normal contacts. The main difference being the contact material which, instead of the Ti/Au bilayer in section 2.3, was chosen to be a Ti/Al/Ti trilayer (14/65/8 nm) inspired by the measurements of in Refs. [22, 74]. In bulk form, aluminum has a superconducting transition temperature of ∼ 1.2 K, and due the proximity effect the lower Ti layer, which makes the actual contact to the nanowire, get an enhanced Tc from its bulk value (400 mK). From the measurements presented below it is found that the contact trilayer has a transition tem56 Closed QD Kondo Physics Conductance Fluctuations 15 G [e2/h] 10 2 1 0 -34 -33 -22 -21 -3 0 3 Gate [V] Figure 2.29: Linear conductance G as a function of gate volgate Vg measured at 300 mK for a nanowire device with superconducting Ti/Al/Ti leads. The inset shows an SEM micrograph of the device. perature Tc ≈ 750 mK. The top Ti-layer is added to protect the aluminum from oxidation. An SEM image of a typical device can be seen in the inset to Fig. 2.29. Since, as in the previous section, Ti makes the actual contact to the nanowire we expect the contact characteristics to be similar. This is indeed the case, as illustrated in Fig. 2.29 which shows the linear conductance as a function of gate voltage measured at 300 mK. The results are qualitatively similar to those presented in section 2.3 and the three regimes treated in the previous sections can again be identified. Below, the influence of the superconductor on the transport characteristics is treated separately for the three regimes. S-Nanowire-S junction in the closed-dot regime In this section we briefly consider the closed-dot regime observed for the lowest gate voltages in Fig. 2.29 where the coupling to the dot is Γ ≈ 150 − 200 µV as determined from the width of the Coulomb peaks, i.e., EC À ∆ ≈ Γ27 . This regime has been addressed experimentally in Refs. [57, 72, 76] for nanotube weak links and in Ref. [71] for metal particles coupled to superconducting leads. Figure 2.30(a) shows a bias spectroscopy plot measured for this region. The familiar pattern of Coulomb diamonds is observed, however, the diamond degeneracy points do not touch at Vsd = 0 V as was the case for normal metals (Fig. 2.16) but are instead separated by an amount ∼ 0.48 mV as indicated by the arrows. Such behavior has been observed before for nanotube weak links[57] and can be qualitatively understood by considering the condition for the onset of quasi particle (QP) tunneling through the resonant level as illustrated in the schematics in Fig. 27 Note, that this estimate is performed while the leads are in the superconducting state since no magnetic field was available to suppress the superconductivity. 57 (a) 3.0 Vsd [mV] 1.5 0.0 -1.5 -3.0 -26.75 -26.50 -26.25 -26.0 Vg [V] (b) (c) (d) (e) ∆ ∆ ∆ ∆ eVsd ∆ Figure 2.30: (a) Bias spectroscopy plot of a nanowire device contacted with superconducting leads in the closed dot regime. The Coulomb diamonds do not come together as indicated by the arrows. This is due to the gap in the superconducting contacts leading to a finite bias threshold Vsd = ±2∆/e for QP tunneling as sketched in (b)-(d); see text. (e) Schematic illustration of the origin of the gate-dependent structure between the diamond tips observed in (a). The sketch illustrates that the process involving one Andreev reflection (n = 1) is only possible when |Vsd | ≥ ∆/e and then only occurs through the resonant level if the gate is adjusted such that αVg = Vsd /2. 2.30(b)-(c). For voltages |Vsd | < 2∆/e no direct quasi-particle tunneling occurs since there are no available states to tunnel into (Fig. 2.30(b)). The threshold is at Vsd = 2∆/e as illustrated in Fig. 2.30(c) and if the position of the level is changed by the gate a larger bias is required (Fig. 2.30(d)). Except for this ±2∆ offset, the condition for QP tunneling is the same as that which determines the diamond edge for a QD with normal leads (section 2.4) and thus the Coulomb diamond pattern appears unaltered except for a ±2∆ gap. Note, that for Tc = 750 mK the value of the gap is 4∆ ∼ 4 ∗ 1.75kB Tc ∼ 460 µ eV in good agreement with the observed value. As seen in Fig. 2.30(a) there is, however, a pronounced highly gate-dependent structure observed for |Vsd | < 2∆/e below the diamond tips, i.e., when the resonant level is in the gap. This structure is interpreted as a signature of Andreev reflections. For gate voltages away from the diamond tips (multiple) AR’s are highly suppressed due to Coulomb blockade and no SGS is observed. Contrary, between the opposing diamond tips, sub-gap peaks are observed when MAR processes can occur through the resonant level. The dependence on the position of the level has been analyzed theoretically in 58 Refs. [73, 82] in the case of negligible Coulomb interaction, and experimentally in Ref. [76] for nanotube quantum dots with superconducting leads. Since our emphasis will be on the more open regime discussed below we will not go into the details of these models. The gate-dependence can, however, be qualitatively understood by considering the condition for the processes involving one AR (n = 2) to occur through the resonant level. In Fig. 2.28(b) the process was illustrated in the non-resonant case occurring at a threshold bias Vsd = ∆/e. The corresponding situation in the case of a resonant junction is illustrated in Fig. 2.30(e). It is seen, that for the level positioned symmetrically between the source and drain, the n = 2 process does not connect through the level and therefore does not contribute significantly to the current. Aligning the level to the Fermi-level of the drain by changing the gate potential αVg = ∆/2e reestablishes the resonance (here α is the scaling parameter for the coupling of the gate to the dot - cf., section 2.4). This constitutes the threshold for the n = 2 process in the resonant junction and for increasing bias the gate must be further adjusted to maintain the level in resonance. Similarly, the threshold of the higher order MAR processes depends on the position of the level resulting in a gate-dependent structure as observed in Fig. 2.30(a). For the rightmost diamond of Fig. 2.30(a) the coupling to the leads is stronger, and cotunneling contributes to the current leading to a faint subgap structure even in middle of the diamond. Below, we discuss the device for the middle region of Fig. 2.29 where the coupling is even stronger and the cotunneling contributes significantly. 59 Influence of Kondo correlations on the subharmonic gap structure of quantum dots Since the discovery of the Kondo effect in quantum dots (QD)[51] this system has received extensive theoretical and experimental attention. As discussed in section 2.5 the effect emerges for QD’s strongly coupled to the leads when the total spin of the electrons on the QD is non-zero e.g., if it hosts an odd number of electrons N . The conduction electrons in the leads then screen the spin through multiple spin-flipping cotunnel events resulting in a correlated many-electron Kondo-state which is experimentally observable as an increased linear conduction through the dot. In the present case where the leads are superconductors the conduction electrons form spinsinglet Cooper-pairs incapable of flipping the dot spin and thus, the Kondo effect and superconductivity constitute competing many-body effects. In the pioneering work by Buitelaar et. al., this competition was experimentally confirmed in investigations of the linear conductance of Kondo-dots in multi-wall carbon nanotubes connected to superconductors. It was indeed found that the Kondo-state persists when the energy needed for breaking the Cooper-pairs are compensated for by the energy gained by forming the Kondo state. Otherwise the Kondo state is suppressed and the Kondoinduced increase in the linear conductance disappears. As discussed above, electron transport for finite bias, smaller than the gap of the superconductors, occurs through the process of Andreev reflections where electrons impinging on a superconductor interface are transferred as Cooper pairs by the reflection of holes. In the previous section these features were suppressed due to Coulomb blockade. In this section we treat nanowire devices in the intermediate regime where the Kondo effect is observed for normal-state leads. No experimental study of the influence of Kondo correlations on the SGS has yet been reported and this is the focus of the present section. We first characterize the device with the contacts in the normal state. Because of the relatively high critical field and the very high g-factor of InAs[30, 44] g ≈ 8 driving the contacts normal with a magnetic field will significantly perturb the Kondo state, and instead we study the device characteristics for temperatures above Tc . Figure 2.31(a) shows the linear conductance G as a function of backgate voltage Vg for temperatures T = 750 − 950 mK when the contacts are normal (solid lines). A series of overlapping Coulomb peaks are observed and the temperature dependencies of the valley conductances Gv (for T > Tc ) are indicated by the arrows. In four valleys the conductance decreases upon lowering the temperature as expected for normal Coulomb blockade, however, for the three valleys labeled κ1 − κ3 the reverse behavior is observed indicating Kondo physics. The left inset shows the stability diagram measured at T = 800 mk showing the familiar pattern of Coulomb diamonds and confirming the presence of a high 60 G [e2/h] 2 0 κ1 -3 (b) 3 800 mk Vsd [mV] 3 Vsd [mV] 3 κ3 b3 0 0.0 -3 Vg 1.5 300 mk κ1 Vg G [e2/h] (a) κ2 b2 0.6 0.0 b2 1 b1 0 -21.5 0.6 κ1 b1 κ1 κ2 -21.25 b3 -21.0 Vg [V] κ3 0.0 -20.75 -0.6 0.0 0.6 Vsd [mV] Figure 2.31: (a) Linear conductance in the Kondo-regime for temperatures T = 750 mK − T = 950 mK (black-red) and at 300 mK (dashed). The qualitative temperature dependence of the valley conductances for temperatures above Tc are indicated by arrows. The three high-conductance valleys κ1 − κ3 exhibit the temperature dependence expected for the Kondo effect. Leftmost inset shows the stability diagram for the region of κ1 at 800 mK clearly showing the zero-bias Kondo ridge (darker = less conductive). At 300 mK, i.e. below Tc , the Kondo ridge has been suppressed (rightmost inset). Instead, a double peak structure is observed as indicated by the arrows (see text). (b) Show dI/dVsd vs. Vsd through the center of all the diamonds for T = 800 mK confirming the presence Kondo peaks in the κ-valleys and the absence of such in the remaining diamonds. conductance Kondo ridge at Vsd = 0 V through the diamond of κ1. Panel (b) confirms the presence of Kondo peaks above Tc in the κ-diamonds and the absence of such in the even-N diamonds. The black dashed line in (a) shows G vs. Vg for T = 300 mK, i.e., below the superconducting transition temperature. Instead of continuing its increase as expected for the Kondo effect (without superconductivity) the valley conductances of κ1−κ3 decrease below their values at 950 mK. This result is consistent with the findings of Ref. [75] and shows that the binding energy ∼ kB TK of the Kondo states κ1 − κ3 are less than the binding energy of the superconducting Cooper pairs ∆ ≈ 1.75kB Tc . Here kB is Boltzmanns constant and TK is the Kondo temperature and thus Tκ1 , Tκ2 , Tκ2 < ∆/kB ≈ 1.3 K (The temperature dependencies are investigated in detail below). The suppression of the Kondo state is confirmed by the disappearance of the Kondo ridge in the stability diagram measured at 300 mK shown for κ1 in the rightmost inset to Fig. 2.31(a). In consistency with Ref. [75] we have also observed Kondo ridges that survive the transition to superconducting contacts and are further enhanced by the superconductor as expected for TK > ∆/kB . Such behavior 61 is demonstrated in Appendix D, but for the present discussion we focus on the case of ”suppressed” Kondo diamonds. The finite bias peaks indicated by the arrows in Fig. 2.31(a) are observed throughout the stability diagram and appear symmetrically around Vsd = 0 V. These are manifestations of the MAR processes as discussed in the previous section, however, in the present case the dot is more strongly coupled and the peaks can be observed also in the middle of the diamonds. The outermost peak are at Vsd = 2∆/e when the peaks in the density of states at the gap edges line up as illustrated in Fig. 2.30(c)28 . The transport for energies below the superconducting energy gap (|Vsd | < 2∆/e)) is mediated by multiple Andreev reflections (MAR) and peaks in dI/dVsd are expected each time a new Andreev process becomes accessible. The stability diagram in Fig. 2.32(a) shows a detailed measurement of these low-bias features for the Coulomb diamonds of κ1 and κ2. As discussed above, a complicated peak structure is observed close to the degeneracy points of the Coulomb diamonds since, in this region, the MAR occur resonantly through the gate-voltage dependent dot level[76, 73]. We restrict our discussion to the middle region of the Coulomb diamonds where the peak positions are gate-independent and transport occurs by cotunneling between the two superconductors. In this case, the MAR peaks at |Vsd | = 2∆/ne, n = 1, 2, . . . are expected to appear with decreasing intensities ∝ tn (t being the effective transparency of the device) due to the strong Coulomb repulsion[83]. The black trace in Fig. 2.32(b) shows the differential conductance along the black dashed line in Fig. 2.32(a) through the middle of an even-N diamond. Indeed, peaks are observed at |Vsd | = 2∆/e with fainter shoulders at |Vsd | = ∆/e. As clearly seen in Fig. 2.32(a) the sub gap structure in the odd-N diamonds of the suppressed Kondo ridges, κ1 and κ2, is clearly modified. Unexpectedly, the peaks at |Vsd | = ∆/e are more than 5 times more intense than the peaks at |Vsd | = 2∆/e. This is emphasized by the red trace in Fig. 2.32(b) which shows a trace through the middle of the κ1-diamond. This contrasts the expectations for simple tunneling between the two superconductors and the presence of the Kondo effect in the normal state points towards electron-electron correlations as the origin of the modified sub gap structure29 . Further support of the connection between the Kondo effect and the enhanced ∆/e-peak is provided in Fig. 2.32(c): The lower panel shows the linear conductance (right axis) for a gate-range of 5 odd-N and 6 even-N Coulomb valleys. In the three odd-N valleys κ1 − κ3 from Fig. 2.31(b) and an additional one, κ0, the Kondo effect was observed in the normal state, and as seen in the upper panel, the enhanced ∆/e-peak is ob28 Ignoring the level for the present discussion. We note, that a similar enhanced ∆/e-peak in the SGS of a carbon nanotube QD was observed in Ref. [75] and its possible connection to the Kondo effect was suggested, however, no conclusions were drawn 29 62 Vsd [mV] 0.3 b1 b2 κ1 b3 κ2 0.0 dI/dVsd (e2/h) (a) 2 1 -0.3 0 -21.5 -21.0 (c) (b) dI/dVsd b2 0.0 Vsd [mV] 0.4 2 0.2 1 ο1 0.0 κ0 0.30 κ1 κ2 -22.0 κ3 G [e2/h] Bias [mV] κ1 e2 h -0.30 Vg [V] -21.25 0 -21.0 Gate [V] Figure 2.32: (a) Bias spectroscopy for the Vg -region of κ1 and κ2 and −0.35 mV ≤ Vsd ≤ 0.35 mV emphasizing the structure around the superconducing gap (darker = less conductive). (b) dI/dVsd vs. Vsd through the middle of two neighboring diamonds (lines in (a)) showing the enhanced peaks at Vsd ≈ ∆ for the Kondo-diamond (curves offset for clarity). (c) Bias spectroscopy for a larger Vg -region showing the enhanced ∆-peak for four Kondo diamonds. Main panel shows the linear conductance (right axis) and the distance between ∆ and 2∆-peaks (left axis). Dashed lines indicate the position of 2∆ and 4∆. 63 served in the SGS of each Coulomb diamond. In the remaining diamonds, including o1 with odd-N (which did not show the Kondo effect in the normal state), the conventional SGS is observed. Thus, the effect seems not to be caused alone by N being odd. In the lower panel of Fig. 2.32(c) the separations of the SGS peaks are extracted for each Coulomb valley (left axis). The separations of the 2∆/e-peaks depend slightly on N and are increased in the κ-valleys with respect to the even-N values30 , however, the enhanced ∆/e-peaks in the κ-valleys always appear at half the separation of the 2∆/e-peak in the same valley. The physical mechanism responsible for the peak at Vsd = ∆/e was schematically illustrated in Fig. 2.28(a): An electron co-tunnels trough the dot and gets reflected at the superconductor interface as a hole with the creation of a Cooper-pair in the superconductor. In the discussion of the MAR structure in the closed dot regime it was noted, in connection with Fig. 2.30(e), that the ∆-peak was only present when the gate was adjusted to tune the level in alignment with the chemical potential of one of the leads. This effect has been experimentally and theoretically established[73, 76]. In the present case the level is far away from the gap but, by a similar reasoning, we note that the enhancement of the ∆-peaks in the κ-diamonds is consistent with the presence of a narrow Kondo-related peak in the density of states at the chemical potentials of the superconductors. In the case of normal leads, this is indeed what is expected for the Kondo effect in the presence of an applied bias; the Kondo peak splits into two components pinned to the chemical potentials of the leads. There are no theoretical works on the finite-bias analog for the case of superconducting leads but a few studies have focused on the equilibrium situation[84, 85, 86] where it is found that correlations do indeed result in a peak within the gap in the spectral function, also for TK < ∆. The position of the peak, however, is found to depend on TK (also for TK > ∆) and some calculations predict the presence of two peaks symmetrically situated in the gap. This contrasts the present measurements which would be consistent with a single peak being situated at the center for all TK . The situation is still under theoretical consideration and the finite bias case remains to be treated. Temperature dependence of the ∆-peak One of the most distinct features of the conventional Kondo effect is provided by the temperature dependence of the valley conductance, and to investigate further the origin of the ∆/e-peak we study the temperature dependence of the sub-gap structure. For T > Tc the increasing Gv for lower temperatures is qualitatively indicated by the arrows in Fig. 2.31(b) for κ1 − κ3. Figure 2.33(b) shows Gv (T ) (solid symbols) for κ1 exhibiting the increase for T & 30 Similar fluctuations were observed in Ref. [75] for nanotube based quantum dots. 64 2 (a) (b) dI/dVsd 310 mK dI/dVsd [e2/h] e 2h 1.15 1.00 0.85 1700 mK -0.5 0.5 0.5 1.0 T [K] 2.0 (c) (d) 1.6 dI/dVsd dI/dVsd [e2/h] 2 e 2h 0.0 Vsd [mV] 1.4 1.2 1.0 -0.2 0.0 Vsd [mV] 0.2 0.3 0.4 0.5 T [K] 0.6 0.7 Tc Figure 2.33: (a), dI/dVsd vs. Vsd through κ1 for different temperatures 310 mK - 1700 mK (offset for clarity). The formation of the ∆/e-peaks can be followed. Symbols correspond to traces in (b) showing the suppression of the valley conductance (dashed line guides the eye) and the increase of the average ∆-peak heights upon cooling below Tc (solid line shows ∆bcs (T ) see text). (c) Measurements as in (a) from another device. The ∆ and 2∆ peaks can be clearly resolved and the average peak heights are extracted in (d) - the 2∆-points offset for clarity. The average 2∆-peak height is well described by the temperature dependence of the gap (solid lines) but the behavior of the 2∆-peak is not captured by ∆bcs (T ) (solid lines) - see text. 1 K followed by a saturation around Tc and a suppression of the conductance when the superconducting gap develops at lower temperatures. To investigate the temperature dependence of the finite-bias peaks in the sub-gap structure, traces of dI/dVsd vs. Vsd were measured through the middle of the Coulomb diamonds for different temperatures. Figure 2.33(a) shows the result for κ1 for temperatures 1700 mK − 310 mK. Upon lowering the temperature, the initial formation of the Kondo peak for T > Tc is observed followed by the formation of the sub-gap peaks at |Vsd | = ∆/e with shoulders at |Vsd | = 2∆/e for T < Tc . As shown on the figure (red dashed lines), the temperature dependence of the peak positions follow the expected temperature dependence of the BCS gap ∆BCS (T ). To allow for a better comparison of the evolution of the ∆ and 2∆ peaks we include in 65 Vsd [mV] * * 0.4 ** * 0 -.4 -5 -4 -3 Gate [V] Figure 2.34: Low bias stability diagram of a nanowire quantum dot with superconducting leads. The bare arrows points to diamonds exhibiting an enhancement of the ∆-peak, starred arrows points to diamonds with a surviving zero-bias Kondo ridge and the double starred arrow indicated diamond of the measurement presented in Fig. 2.33(c) and (d). The slight bias drift evident in (b) is due to a changing optocoupler-offset during the course of the measurement. panel (c) a similar measurement performed for a suppressed Kondo ridge in a different device where the four peaks can be clearly distinguished. In panels (b) and (d) the temperature dependence of the peak heights and the valley conductances have been extracted. In both cases the ∆-peaks increases monotonically whereas the 2∆-peaks of the measurement in panel (c) saturates below ∼ 0.5 K. Since the temperature dependence is governed by the temperature dependence of the gap we expect the measured peak heights31 (in the absence of correlations) to be proportional to ∆(T )[87, 88]. Such a fit is made in panel (d) and indeed describes the behavior of the 2∆peak very well. For the ∆-peak, however, the continuing strengthening for lower temperatures is not well reproduced by ∆(T ), thereby adding further evidence to the importance of correlations for the origin of this peak. Finally, we note that the observed phenomenon is not limited to the four odd-N diamonds presented here. Figure 2.34 shows a low-bias stability diagram for another device exhibiting the same enhancement of the ∆-peak in 11 diamonds. Altogether we have observed the increased ∆/e peak for ∼ 20 suppressed Kondo ridges in two different nanowire devices and the same phenomenon has been observed in single-wall carbon nanotube-based devices fabricated and measured by K. Grove-Rasmussen and H.I. Jørgensen. In appendix D an example from such a device is presented. This shows that the phenomenon is general for Kondo-dots connected to superconducting leads. More work is needed, however, to fully understand its origin. 31 The measured conductance peaks are proportional to the magnitude of the currentjumps at eVsd = 2∆/n in the IV-curves. 66 S-Nanowire-S junctions in the open regime We now briefly turn to the S-Nanowire-S junction in the rightmost gate range of Fig. 2.29 displaying the characteristics of a diffusive conductor as treated in section 2.6 for the case of normal metal leads. Measurements on devices in the open regime have been reported in Ref. [89] for diffusive multi-wall carbon nanotube devices, in Refs. [77, 78] for single-wall carbon nanotube devices in the Fabry-Perot regime, and in Refs. [22, 23] for semiconducting InAs and Si/Ge nanowires, respectively. As mentioned earlier our investigation of the superconducting devices has been focused on the Kondo regime discussed in the previous section and only a few measurements have been performed in this regime. This section is included, however, for the sake of completeness. Figure 2.35(a) shows a bias spectroscopy plot of the device in the open regime. For |Vsd | & 0.3 mV the pattern interpreted in section 2.6 as caused by quantum interference (UCF) is again recognized. As expected. in view of the preceding sections, a pronounced structure is observed at low bias due to the superconductors. Panel (b) shows a detailed bias-spectroscopy measurement of this sub-gap structure32 . Since in this gate range, Γ À Uc there is no suppression of MAR due to the Coulomb interaction and the peaks due to quasi particle tunneling are clearly observed in the whole gate range for Vsd ≈ ±190 µV. Higher order structures can be identified as well, as indicated by the arrows in panel (c). The lower value of 2∆ in these measurements as compared to those discussed in the preceding section might be due to local heating which is expected to be more important for these measurements due to the larger conductance (i.e. higher current). It is clear from the extracted sections in panel (c) that the relative strength of the different MAR processes depends on the transmission of the device, ranging from enhancement to suppression of the higher order processes. A similar behavior has been observed in carbon nanotube based devices[89, 57]. A more systematic investigation might be pursued in a future nanowire study, but the present data does not allow for such an analysis. An interesting feature evident in the measurements of Fig. 2.35(b) and (c) is the narrow (FWHM ∼ 10 − 14 µV) zero-bias peak observed for all gate voltages. Recently, a zero bias dissipationless (super) current was observed in nanowire devices in the open regime [22, 23]. This would correspond to a divergent conductance peak at Vsd = 0 mV. The supercurrent is, however, very sensitive to noise and in Refs. [22, 23] the cryostats were equipped with special (cold) copper-powder filters in order to observe the supercurrent. In the absence of such precautions (as is the case for the present measurements) 32 The data of Fig. 2.35(a) and (b) were measured with a few days and a few 3 He condensation-cycles in between, and despite the overlapping gate interval there is not complete correspondence between the two probably due to a reconfiguration of the charges in the substrate in the meantime. 67 the supercurrent will be smeared and thus, it is tempting to attribute the zero-bias peak in our measurements to a noise-smeared super-current peak. However, direct observation of the dissipationless current is required to prove this. Vsd [mV] (a) 2.5 0.0 -2.5 6 7 8 Vg [V] 10 (c) 7.0 dI/dV [e2/h] Vsd [mV] (b) 0.3 9 0.0 7 (ii) (i) 3.0 iii ii i -0.3 (iii) 5.0 8 9 10 Vg [V] -3 0 Vsd [mV] 3 Figure 2.35: (a) Bias spectroscopy plot (darker = less conductive) of a nanowire device contacted with superconducting leads in the open regime exhibiting quantum interference. At |Vsd | = 2∆/e an enhanced conductivity is observed due to the enhancement of the quasiparticle transport by the peaks in the density of states. Below this bias threshold transport occur by the process of Andreev reflections giving rise to the sub-gap structure emphasized by the detailed low-bias spectroscopy plot in (b) and the sections extracted in (c). The peak at Vsd = 0 mV observed in (b) and (c) for all Vg is interpreted as a smeared super-current peak. The discontinuities in (a) at Vg ∼ 9.1 V and Vg ∼ 9.4 V are due to gate-switches and the slight bias drift evident in (b) is due to a changing optocoupler-offset during the course of the measurement. 68 2.8 Conclusion of nanowire studies In conclusion, we have in this chapter investigated the electrical transport characteristics of devices fabricated from semiconducting InAs nanowires. A fabrication scheme has been developed which has allowed a reasonable fraction of devices to reach the state of low-temperature measurements (despite extensive and obstinate handling problems due to extreme device sensitivity to static electricity). Data has been obtained on nanowires contacted by normal-metal (Ti/Au) leads and the transport characteristics have been investigated in regimes of different barrier transparencies. These investigations were enabled by a tunability of the barriers with the backgate potential. Three transport regimes were treated separately; the Coulomb blockade regime, the Kondo regime, and the open regime. In the CB regime un-tunable devices ”born” with large contact barriers, due to a slight variation in the fabrication scheme, were most extensively studied. A regular CB structure was found, persisting at least over the gate range corresponding to the addition of ∼ 150 electrons. Results of non-linear transport measurements were presented and the level structure was analyzed in details giving charging energies EC ∼ 2.5 meV and average level spacings ∆Eavg ≈ 0.2 meV. By investigating the influence of electrode separation on the back-gate capacitance the confinement of the dot was shown to be defined by the electrodes rather than random defects along the wire. For device with L = 300 nm electrode spacing, the capacitance of the dot corresponded to an effective length of ∼ 65 nm for the wire segment comprising the dot. For larger barrier transparencies, evidences of the higher order transport processes of cotunneling was found. For Coulomb diamonds corresponding to the dot hosting an odd number of electrons, the Kondo effect was observed. Being the first observation of the Kondo effect in nanowire systems its characteristics were studied in detail and found to agree with the conventional theory for quantum dots. The splitting of the Kondo ridge in an external perpendicular magnetic field was found to be a convenient way of determining the Lande g-factor of the devices and g ≈ 8 was found for all Kondo ridges and devices. The value is considerably down-shifted with respect to the bulk value but agrees with reported studies of the g-factor in confined InAs structures. The observation of the excited state structure and the Kondo effect in the nanowire quantum dots show unambiguously that these relatively simple devices constitute well behaved and coherent quantum devices at low temperatures. For even higher transparencies the devices acted as coherent diffusive conductors and electron interference leads to the phenomena of (universal) conductance fluctuations when varying the external parameters of gate, magnetic field and bias. In the magneto conductance, weak (anti) localization was found and by averaging over gate potential while varying the magnetic 69 field the weak localization correction to the conduction could be separated from the random fluctuations enabling an estimate of the elastic scattering length le ≈ 100 nm, the spin-orbit scattering length lso ≈ 200 nm and the phase-coherence length lϕ ≈ 220 nm (at 4.2 K). Nanowire devices contacted by superconducing leads were realized following the same fabrication scheme but choosing a (Ti/Al/Ti) trilayer for contacting which turns superconducting below Tc ≈ 750 mK. The effects of the superconducting leads were studied in the three transport regimes, however, with emphasis on the intermediate regime where the conventional Kondo effect was observed for temperatures above Tc . Upon cooling the devices below Tc , the fate of the Kondo state depends on its binding energy with respect to the binding energy of the Cooper-pairs of the superconductor and in most cases the Kondo ridge was suppressed. Nevertheless, Kondocorrelations were found to have a pronounced influence on the subharmonic gap structure, greatly enhancing the ∆-peak in the differential conductance. The same effect has also been observed by K. Grove-Rasmussen and H.I. Jørgensen in our group, for quantum dots in carbon nanotubes contacted by superconductors, confirming the general character of the phenomenon. The measurements are consistent with a Kondo-related peak situated at each chemical potential, as is the case for the finite-bias situation for Kondo-dots with normal leads, however, no theoretical model has yet been solved to substantiate this further. These issues remain under theoretical consideration. In the closed-dot regime the main effect of the superconductor was to open a gap in the stability-diagram and in the open regime the effects of multiple Andreev reflections were observed and indications of a supercurrent were found but the details remain for future studies. All the results presented in this chapter have been observed in more than one device. Outlook As discussed in the introduction, the nanowire system in general, provides a large range of possibilities for nanoscale research and applications. For electrical device applications alone, countless combinations exist of possible nanowire crystals (with various properties), the choice of contact materials and the possibilities for tailored heterostructures (both radial and axial). In this respect the nanowire system is unique and is still in its infancy and therefore holds plenty of room for significant scientific discoveries. There are a number of obvious extensions of the experiments presented in this thesis. Useful information for quantifying possible differences between the MBE-grown wires of this project and those synthesized by ”conventional” techniques could be extracted from a systematic study of the magnetoconductance measurements of weak localization (cf. Ref. [36]). Such experiments should be carried out in the near future. Fabrication wise, fol70 lowing the recent reports in Refs. [20, 22, 90] on local electrostatic control of the quantum dot barriers, would allow extended flexibility and the possibility of defining multiple dots. Such systems are interesting for quantum information and recent theoretical proposals for electrical control over the quantum dot spin via the spin-orbit interaction in InAs nanowires provide an exciting perspective[91]. Also in extension of the measurements on nanowires contacted by superconductors, the development of local gate control would allow exciting investigations of the possibility of forcing the individual electrons of a Cooper-pair through two spatially separated dots while still preserving their spin entanglement[92]. The possibilities and perspectives are great and it will be exciting to follow the future developments. 71 Bibliography [1] A. Morales and C. Lieber, “A laser ablation method for the synthesis of crystalline semiconductor nanowires”, Science 279(5348), 208 (1998). [2] L. Samuelson, C. Thelander, M. Björk, M. Borgstrom, K. Deppert, et al., “Semiconductor nanowires for 0D and 1D physics and applications”, Physica E 25(2-3), 313 (2004). [3] R. Wagner and W. Ellis, “Vapor-liquid-solid mechanism of single crystal growth”, Appl. Phys. Lett. 4(5), 89 (1964). [4] A. Persson, M. Larsson, S. Stenstrom, B. Ohlsson, L. Samuelson, and L. Wallenberg, “Solid-phase diffusion mechanism for GaAs nanowire growth”, Nat. Mater. 3(10), 677 (2004). [5] D. Milliron, S. Hughes, Y. Cui, L. Manna, J. Li, L. Wang, and A. Alivisatos, “Colloidal nanocrystal heterostructures with linear and branched topology”, Nature 430(6996), 190 (2004). [6] X. Duan and C. Lieber, “General synthesis of compound semiconductor nanowires”, Adv. Mat. 12(4), 298 (2000). [7] B. Ohlsson, M. Björk, M. Magnusson, K. Deppert, L. Samuelson, and L. Wallenberg, “Size-, shape-, and position-controlled GaAs nanowhiskers”, Appl. Phys. Lett. 79(20), 3335 (2001). [8] L. Jensen, M. Björk, S. Jeppesen, A. Persson, B. Ohlsson, and L. Samuelson, “Role of surface diffusion in chemical beam epitaxy of InAs nanowires”, Nano Lett. 4(10), 1961 (2004). [9] Y. Wu, Y. Cui, L. Huynh, C. Barrelet, D. Bell, and C. Lieber, “Controlled growth and structures of molecular-scale silicon nanowires”, Nano Lett. 4(3), 433 (2004). [10] Z. Zhong, Y. Fang, W. Lu, and C. Lieber, “Coherent single charge transport in molecular-scale silicon nanowires”, Nano Lett. 5(6), 1143 (2005). [11] K. Hiruma, M. Yazawa, K. Haraguchi, K. Ogawa, T. Katsuyama, M. Koguchi, and H. Kakibayashi, “GaAs freestanding quantum-size wires”, J. Appl. Phys 74(5), 3162 (1993). [12] Z. Wu, X. Mei, D. Kim, M. Blumin, and H. Ruda, “Growth of Aucatalyzed ordered GaAs nanowire arrays by molecular-beam epitaxy”, Appl. Phys. Lett. 81(27), 5177 (2002). 72 [13] M. Gudiksen, L. Lauhon, J. Wang, D. Smith, and C. Lieber, “Growth of nanowire superlattice structures for nanoscale photonics and electronics”, Nature 415(6872), 617 (2002). [14] M. Björk, B. Ohlsson, C. Thelander, A. Persson, K. Deppert, L. Wallenberg, and L. Samuelson, “Nanowire resonant tunneling diodes”, Appl. Phys. Lett. 81(23), 4458 (2002). [15] L. Lauhon, M. Gudiksen, C. Wang, and C. Lieber, “Epitaxial core-shell and core-multishell nanowire heterostructures”, Nature 420(6911), 57 (2002). [16] M. Björk, B. Ohlsson, T. Sass, A. Persson, C. Thelander, et al., “Onedimensional steeplechase for electrons realized”, Nano Lett. 2(2), 87 (2002). [17] M. Mcalpine, R. Friedman, S. Jin, K. Lin, W. Wang, and C. Lieber, “High-performance nanowire electronics and photonics on glass and plastic substrates”, Nano Lett. 3(11), 1531 (2003). [18] Y. Li, F. Qian, J. Xiang, and C. Lieber, “Nanowire electronic and optoelectronic devices”, Mat. Today 9(10), 18 (2006). [19] M. T. Björk, C. Thelander, A. E. Hansen, L. E. Jensen, M. W. Larsson, and L. Wallenberg, L. R. and, “Few-electron quantum dots in nanowires”, Nano Lett. 4(9), 1621 (2004). [20] C. Fasth, A. Fuhrer, M. Björk, and L. Samuelson, “Tunable double quantum dots in InAs nanowires defined by local gate electrodes”, Nano Lett. 5(7), 1487 (2005). [21] J. van Dam, Y. Nazarov, E. Bakkers, S. de Franceschi, and L. Kouwenhoven, “Supercurrent reversal in quantum dots”, Nature 442(7103), 667 (2006). [22] Y. J. Doh, J. A. van Dam, A. L. Roest, E. P. A. M. Bakkers, L. P. Kouwenhoven, and S. de Franceschi, “Tunable supercurrent through semiconductor nanowires”, Science 309(5732), 272 (2005). [23] J. Xiang, A. Vidan, M. Tinkham, R. Westervelt, and C. Lieber, “Ge/Si nanowire mesoscopic Josephson junctions”, Nat. Nanotechnol. 1(3), 208 (2006). [24] S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, 1995). [25] H. Bruus and K. Flensberg, Many-Body Quantum Theory in Condensed Matter Physics (Oxford University Press, 2004). 73 [26] C. Thelander, M. Björk, M. Larsson, A. Hansen, L. Wallenberg, and L. Samuelson, “Electron transport in InAs nanowires and heterostructure nanowire devices”, Solid State Commun. 131(9-10), 573 (2004). [27] M. Aagesen, et. al., in preparation. [28] M. Aagesen, ph. D. thesis, University of Copenhagen, in preparation. [29] J. Johansson, L. Karlsson, C. Svensson, T. Martensson, B. Wacaser, et al., “Structural properties of (111)b-oriented III-V nanowires”, Nat. Mater. 5(7), 574 (2006). [30] M. T. Björk, A. Fuhrer, A. E. Hansen, M. W. Larsson, L. E. Froberg, and L. Samuelson, “Tunable effective g factor in InAs nanowire quantum dots”, Phys. Rev. B 72(20), 201 307 (2005). [31] M. Larsson, J. Wagner, M. Wallin, P. Hakansson, L. Froberg, L. Samuelson, and L. Wallenberg, “Strain mapping in free-standing heterostructured wurtzite InAs/InP nanowires”, Nanotechnology 18, 015 504 (2006). [32] Z. Zanolli and U. von Barth, “All-electron study of InAs and GaAs wurtzite: structural and electronic properties”, cond-mat/0610066. [33] Z. Zanolli, F. Fuchs, J. Furthmueller, U. von Barth, and F. Bechstedt, “GW band structure of InAs and GaAs in the wurtzite phase”, condmat/0609616. [34] C. Kittel, Introduction to Solid State Physics (John Wiley & Sons, Inc. 2005). [35] S. Bhargava, H. Blank, V. Narayanamurti, and H. Kroemer, “Fermilevel pinning position at the Au-InAs interface determined using ballistic electron emission microscopy”, Appl. Phys. Lett. 70(6), 759 (1997). [36] A. Hansen, M. Björk, C. Fasth, C. Thelander, and L. Samuelson, “Spin relaxation in InAs nanowires studied by tunable weak antilocalization”, Phys. Rev. B 71(20), 205 328 (2005). [37] C. Fasth, A. Fuhrer, L. Samuelson, V. Golovach, and D. Loss, “Direct measurement of the spin-orbit interaction in a two-electron InAs nanowire quantum dot”, cond-mat/0701161. [38] N. Balshaw, Practical Cryogenics (Oxford Instruments (UK) Ltd, Scientific Research Division, 1996). [39] Y. Huang, X. Duan, and C. Lieber, “Nanowires for integrated multicolor nanophotonics”, Small 1(1), 142 (2005). 74 [40] L. Kouwenhoven, C. Marcus, P. McEuen, S. Tarucha, R. Westervelt, and N. Wingreen, “Electron transport in quantum dots”, in Mesoscopic Electron Transport, edited by L. Sohn, L. Kouwenhoven, and G. Schön (Proceedings of the NATO Advanced Study Institute on Mesoscopic Electron Transport, 1997), p. 105. [41] H. van Houten, C. Beenakker, and A. Starring, Single Charge Tunneling (New York: Plemnum Press, 1991). [42] C. Beenakker and H. van Houten, “Quantum transport in semiconductor nanostructures”, Solid State Phys. 44, 1 (1991). [43] J. Nygård, Experiments on Mesoscopic Electron Transport in Carbon Nanotubes, Ph.D. thesis, University Of Copenhagen (2000). [44] T. Jespersen, M. Aagesen, C. Sørensen, P. Lindelof, and J. Nygård, “Kondo physics in tunable semiconducting nanowire quantum dots”, Phys. Rev. B 74, 233 304 (2006). [45] S. de Franceschi, S. Sasaki, J. Elzerman, W. van der Wiel, S. Tarucha, and L. Kouwenhoven, “Electron cotunneling in a semiconductor quantum dot”, Phys. Rev. Lett. 86(5), 878 (2001). [46] J. Kondo, “Resistance minimum in dilute magnetic alloys”, Prog. Theor. Phys. 32(1), 37 (1964). [47] L. Kouwenhoven and L. Glazman, “Revival of the Kondo effect”, Phys. World 14(1), 33 (2001). [48] A. Hewson, The Kondo problem to heavy fermions (Cambridge University Press, 1997). [49] L.I. Glazman, M. Pustilnik, Low-temperature transport through a quantum dot, H. Bouchiat et. al., eds. (Lectures notes of the Les Houches Summer School 2004), pp. 427-478. cond-mat/0501007. [50] A. Kogan, S. Amasha, D. Goldhaber-Gordon, G. Granger, M. Kastner, and H. Shtrikman, “Measurements of Kondo and spin splitting in singleelectron transistors”, Phys. Rev. Lett. 93(16), 166 602 (2004). [51] D. Goldhaber-Gordon, H. Shtrikman, D. Mahalu, D. Abusch-Magder, U. Meirav, and M. Kastner, “Kondo effect in a single-electron transistor”, Nature 391(6663), 156 (1998). [52] J. Nygård, D. Cobden, and P. Lindelof, “Kondo physics in carbon nanotubes”, Nature 408(6810), 342 (2000). 75 [53] M. Buitelaar, A. Bachtold, T. Nussbaumer, M. Iqbal, and C. Schönenberger, “Multiwall carbon nanotubes as quantum dots”, Phys. Rev. Lett. 88(15), 156 801 (2002). [54] J. Park, A. Pasupathy, J. Goldsmith, C. Chang, Y. Yaish, et al., “Coulomb blockade and the Kondo effect in single-atom transistors”, Nature 417(6890), 722 (2002). [55] W. Liang, M. Shores, M. Bockrath, J. Long, and H. Park, “Kondo resonance in a single-molecule transistor”, Nature 417(6890), 725 (2002). [56] P. Utko, J. Nygård, M. Monthioux, and L. Noe, “Sub-kelvin transport spectroscopy of fullerene peapod quantum dots”, Appl. Phys. Lett. 89(23), 233 118 (2006). [57] K. Grove-Rasmussen, Electronic Transport in Single Wall Carbon Nanotubes, Ph.D. thesis, University Of Copenhagen (2006). [58] S. Cronenwett, Coherence, charging, and spin effects in quantum dots and point contacts, Ph.D. thesis, Harvard University (2001). [59] D. H. Cobden, M. Bockrath, P. L. Mceuen, A. G. Rinzler, and R. E. Smalley, “Spin splitting and even-odd effects in carbon nanotubes”, Phys. Rev. Lett. 81(3), 681 (1998). [60] S. Cronenwett, T. Oosterkamp, and L. Kouwenhoven, “A tunable Kondo effect in quantum dots”, Science 281(5376), 540 (1998). [61] D. Goldhaber-Gordon, J. Gores, M. Kastner, H. Shtrikman, D. Mahalu, and U. Meirav, “From the Kondo regime to the mixed-valence regime in a single-electron transistor”, Phys. Rev. Lett. 81(23), 5225 (1998). [62] J. Paaske, A. Rosch, and P. Wolfle, “Nonequilibrium transport through a Kondo dot in a magnetic field: Perturbation theory”, Phys. Rev. B 69(15), 155 330 (2004). [63] A. Hewson, J. Bauer, and W. Koller, “Field dependent quasiparticles in a strongly correlated local system”, Phys. Rev. B 73(4), 045 117 (2006). [64] A. Kiselev, E. Ivchenko, and U. Rossler, “Electron g factor in one- and zero-dimensional semiconductor nanostructures”, Phys. Rev. B 58(24), 16 353 (1998). [65] J. Paaske, A. Rosch, P. Wolfle, N. Mason, C. Marcus, and J. Nygård, “Non-equilibrium singlet-triplet Kondo effect in carbon nanotubes”, Nat. Phys. 2, 460 (2006). 76 [66] W. Liang, M. Bockrath, D. Bozovic, J. Hafner, M. Tinkham, and H. Park, “Fabry-Perot interference in a nanotube electron waveguide”, Nature 411(6838), 665 (2001). [67] J. Nygård and D. Cobden, “Transport phenomena in nanotube quantum dots from strong to weak confinement”, T. Martin et al (eds), EDP Sciences,(2001), cond-mat/0105289. [68] G. Bergmann, “Weak localization in thin-films - a time-of-flight experiment with conduction electrons”, Phys. Rep. 107(1), 1 (1984). [69] S. Sahoo, T. Kontos, J. Furer, C. Hoffmann, M. Graber, A. Cottet, and C. Schönenberger, “Electric field control of spin transport”, Nat. Phys. 1(2), 99 (2005). [70] S. Datta and B. Das, “Electronic analog of the electrooptic modulator”, Appl. Phys. Lett. 56(7), 665 (1990). [71] D. Ralph, C. Black, and M. Tinkham, “Spectroscopic measurements of discrete electronic states in single metal particles”, Phys. Rev. Lett. 74(16), 3241 (1995). [72] V. Krstic, S. Roth, M. Burghard, J. Weis, and K. Kern, “Suppression of superconductor quasiparticle tunneling into single-walled carbon nanotubes”, Phys. Rev. B 68(20), 205 402 (2003). [73] G. Johansson, E. N. Bratus, V. S. Shumeiko, and G. Wendin, “Resonant multiple Andreev reflections in mesoscopic superconducting junctions”, Phys. Rev. B 60(2), 1382 (1999). [74] K. Grove-Rasmussen, H. I. Jørgensen, and P. E. Lindelof, “Interplay between supercurrent and Kondo effect in single wall carbon nanotube Josephson junctions”, cond-mat/0601371. [75] M. Buitelaar, T. Nussbaumer, and C. Schönenberger, “Quantum dot in the Kondo regime coupled to superconductors”, Phys. Rev. Lett. 89(25), 256 801 (2002). [76] M. Buitelaar, W. Belzig, T. Nussbaumer, B. Babic, C. Bruder, and C. Schönenberger, “Multiple Andreev reflections in a carbon nanotube quantum dot”, Phys. Rev. Lett. 91(5), 057 005 (2003). [77] P. Jarillo-Herrero, J. van dam, and L. Kouwenhoven, “Quantum supercurrent transistors in carbon nanotubes”, Nature 439(7079), 953 (2006). [78] H. Jørgensen, K. Grove-Rasmussen, T. Novotny, K. Flensberg, and P. Lindelof, “Electron transport in single-wall carbon nanotube weak 77 links in the Fabry-Perot regime”, Phys. Rev. Lett. 96(20), 207 003 (2006). [79] K. Likharev, “Superconducting weak links”, Rev. Mod. Phys. 51(1), 101 (1979). [80] M. Tinkham, Introduction to superconductivity (Dover Publications, Inc. Mineola, New York, 2004). [81] B. Josephson, “Possible new effects in superconductive tunnelling”, Phys. Lett. 1(7), 251 (1962). [82] A. L. Yeyati, J. C. Cuevas, A. Lopezdavalos, and A. Martinrodero, “Resonant tunneling through a small quantum dot coupled to superconducting leads”, Phys. Rev. B 55(10) (1997). [83] Y. Avishai, A. Golub, and A. Zaikin, “Superconductor-quantum dotsuperconductor junction in the Kondo regime”, Phys. Rev. B 67(4), 041 301 (2003). [84] A. Clerk and V. Ambegaokar, “Loss of pi-junction behavior in an interacting impurity Josephson junction”, Phys. Rev. B 61(13), 9109 (2000). [85] G. Sellier, T. Kopp, J. Kroha, and Y. S. Barash, “pi junction behavior and Andreev bound states in Kondo quantum dots with superconducting leads”, Phys. Rev. B 72(17), 174 502 (2005). [86] J. Lim and M. Choi, “Andreev bound states in the Kondo quantum dots coupled to superconducting leads”, cond-mat/0609333. [87] E. Wolf, Principles of Electron Tunneling Spectroscopy (Oxford University Press, 1985), p. 113. [88] J. Cuevas, A. Martinrodero, and A. Yeyati, “Hamiltonian approach to the transport properties of superconducting quantum point”, Phys. Rev. B 54(10), 7366 (1996). [89] M. Buitelaar, Electron Transport in Multiwall Carbon Nanotubes, Ph.D. thesis, University of Basel (2004). [90] A. Pfund, I. Shorubalko, R. Leturcq, and K. Ensslin, “Top-gate defined double quantum dots in InAs nanowires”, Appl. Phys. Lett. 89, 252 106 (2006). [91] C. Flindt, A. Sørensen, and K. Flensberg, “Spin-orbit mediated control of spin qubits”, Phys. Rev. Lett. 97(24), 240 501 (2006). [92] P. Recher, E. Sukhorukov, and D. Loss, “Andreev tunneling, coulomb blockade, and resonant transport of nonlocal spin-entangled electrons”, Phys. Rev. B 63(16), 165 314 (2001). 78 Chapter 3 EFM on Carbon Nanotubes 3.1 Introduction Figure 3.1(a) shows a schematic illustration of the crystal structure of a single wall carbon nanotube (SWNT). SWNT’s can be viewed as the structure which emerges when rolling a 2D graphene layer into a seamless cylinder. Due to the properties of the underlying graphene structure and the quantization of the electronic wave function around the circumferential direction of the tube, SWNT’s acquire remarkable mechanical, optical, and electrical properties[1] and ever since their discovery in 1991[2] they have received enormous attention from almost every corner of the nano science community. Electrically, SWNT’s can be either metallic or semiconducting depending on the exact crystal structure and they constitute a near-ideal realization of a one-dimensional system. Therefore, many groups have focused on making electrical measurements on carbon nanotubes (CNT) for fundamental as well as applied science. Almost all CNT-based electrical devices share the same starting-point geometry as the nanowire devices discussed in the previous chapter. This geometry is schematically illustrated in Fig. 3.1(b). CNT’s are supported on a thin layer of insulating oxide (e.g., silicon dioxide) which caps a conducting substrate (e.g., highly doped silicon) that can act as a global electrode for electrostatic gating. The CNT’s are usually deposited on the substrate from a suspension in an organic solvent or directly grown on the substrate by chemical vapor deposition (CVD). The electrical contact to the CNT’s is realized by metal electrodes which are usually fabricated by standard top-down lithographic techniques. For many purposes it is crucial to be able to image the CNT distribution prior to electrode fabrication. For instance, the exact positions of the CNT’s on the substrate are often required for single-nanotube devices, and for devices made by evaporation of random contacts it is useful to know the density of CNT’s. Two standard techniques are capable of imaging CNT’s in this geometry. Scanning electron microscopy (SEM) and 79 (a) (b) CNT SiO2 doped Si Figure 3.1: (a) Side-view schematic of the atomic configuration of a section of a carbon nanotube. The crystal structure of this particular tube is denoted (7, 7). (b) Side-view schematic of the typical starting point geometry for CNT-based electrical devices: The CNT is supported on a conducting wafer (e.g. doped Si) which is capped with ∼ 500 nm of insulator (e.g. SiO2 ). The CNT’s studied in this chapter are all in this geometry. tapping-mode atomic force microscopy (AFM). SEM has a number of major drawbacks. Due to the very small size of CNT’s as well as charge build-up in the insulating substrate it is often difficult to achieve sufficient resolution to image CNT’s by SEM. Furthermore, carbon contaminants are often deposited by the scanning beam (thus questioning subsequent use of the imaged tubes) and, more importantly, it has been shown that even low energy SEM induces defects in CNT’s greatly affecting device performance[3, 4, 5]. Tapping mode AFM is (currently believed to be) a non-invasive technique capable of imaging surfaces with very high resolution and it has been widely employed for imaging CNT samples. The technique is, however, relatively slow and if SWNT’s having diameters of ∼ 1 − 3 nm are to be clearly resolved, the image-area is limited to ∼ 20 × 20 µm, making assessment of the larger scale distribution of SWNT’s exceedingly time consuming. In this section our explorations of a relatively new technique for assessing CNT’s in the aforementioned geometry are described. The technique: Electrostatic force microscopy (EFM) employs an atomic force microscope equipped with a conducting cantilever for measuring the electrostatic interaction between the tip and the sample. Since SWNT’s are highly conducting, compared to their insulating support, they are easily resolved with the EFM technique. As EFM makes use of the Coulomb interaction, of longer range than the van der Waals interaction of standard AFM, much faster and larger scans can be reliably carried out without loosing important information. The technique has previously been employed for the study of charging effects in carbon nanotubes[6], for comparison of the conductivity of low-dimensional objects (thereby establishing the insulating nature of λ-DNA)[7, 8] and even for assessment of the density of states of individual CNT’s [9]. The following sections initially provides a brief introduction to the properties of SWNT’s followed by a description of the principle of EFM, the experimental details and the details of the sample preparation. The technique is subsequently employed for 80 • characterization of the large scale distribution of SWNT’s, • assessing the quality of SWNT’s by investigating defects on a single nanotube level, • demonstrating the interplay between SWNT’s and surface charges, and • mapping SWNT’s suspended in a polymer matrix (a highly relevant geometry as will be described below) Most of these results are presented in Refs. [10, 11, 12, 13]. Briefly about carbon nanotubes In this section we briefly review the properties and terminology of singlewall carbon nanotubes relevant for understanding the EFM measurements in the following sections. Already Refs. [1, 14] provide excellent general reviews of SWNT-properties, and detailed accounts of electronic properties and electron transport can (among other places) be found in Refs. [15, 16]. For details on the mechanical and vibrational properties Refs. [17, 14] are good sources. Since detailed knowledge is not needed for our purpose and since most readers will, anyway, most likely be familiar with the basic theory of SWNT’s I simply list below the relevant properties and refer to the literature for the details • Geometric properties. As noted above, single wall carbon nanotubes can be imaginatively formed by rolling a strip of graphene into a seamless cylinder. Obviously there are many different ways of choosing an appropriate strip - leading to tubes of different diameters and symmetries. Figure 3.2(a) shows four examples of different SWNT’s having approximately the same diameters dt ∼ 1.4 nm but different symmetries. The crystal structure of a SWNT can be uniquely identified by the chiral vector C which describes the vector in the underlying 2D graphene which needs to be rolled end-to-end to form the particular tube. Thus, as indicated on the figure, a tube can be designated by the pair of integers (n, m) used in writing C = na1 + ma2 as a linear combination of the graphene unit vectors a1,2 . The special case of tubes with indices (n, n) or (n, 0) have chiral vectors coinciding with a symmetry axis of the graphene and they attain special symmetries themselves - they are non-chiral and are designated armchair and zigzag, respectively. Typically, SWNT’s come with lengths up to a few micrometers, however, occasionally tubes with lengths of hundreds of micrometers are observed1 (section 3.2). The diameters of SWNT’s produced in the 1 The current record is ∼ 4 cm [19]. 81 (a) (12,7) (13,7) (10,10) (b) (15,0) (d) (e) (a) (b) (c) (c) Figure 3.2: (a) 3D illustrations of the structures of non chiral (10, 10),(15, 0) and chiral (12, 7),(13, 7) SWNT’s. The (12, 7) and (13, 7) are almost identical to the naked eye but have different electronic properties. (b),(c),(d) High-resolution TEM micrographs showing individual single and multi wall carbon nanotubes and a section of a bundle of SWNT’s respectively (from [18]). A bundle of (10, 10) SWNT’s is schematically illustrated in (e); in reality the SWNT’s comprising are not identical. 82 lab usually fall in the range 0.7 nm . dt . 3 nm; thinner SWNT’s are energetically unfavorable due to the large strain and only form by use of special growth methods[20] and thicker SWNT’s collapse since graphene is rather soft towards out-of-plane deformations2 . Figure 3.2(b),(c) show high-resolution TEM micrographs of individual single and multi-wall carbon nanotubes respectively. Often, SWNT’s come in the form of thin bundles/ropes comprising a few tens of individual SWNT’s. Figure 3.2(d) shows a TEM micrograph of a section of such a rope and panel (e) shows schematically the tube arrangement in a rope of (10, 10) tubes. • Electronic properties The wave-function of an electron in a SWNT is obviously exposed to periodic boundary conditions in the circumferential direction, and thus the momentum will we be quantized in this direction. For an infinitely long tube the parallel momentum is continues and the band structure can be obtained from the electronic structure of graphene by slicing the graphene dispersion along a set of lines parallel to the tube axis. Graphene is a semi-metal and the valence band touches the conduction band at the discrete corner points (K-points) of the hexagonal Brillouin zone. Thus, a particular (n, m)nanotube will be metallic only if the allowed lines in k-space happen to slice through a K-point. Otherwise, it will be semiconducting. The number and orientation of the lines on the underlying graphene dispersion depend on the tube indices (n, m) and it can be shown that it will be metallic if n−m is divisible by 3 [1, 14]. Therefore, even though they appear almost identical to the eye, the (12, 7) tube of Fig. 3.2(a) is semiconducting whereas the (13, 7) tube is metallic. There is much more to be said about the electronic structure of SWNT’s especially if considering (low-temperature) electronic transport, however, for the purpose of the EFM studies, this is all that is required, and the reader is referred to the literature for further details. These points summarize what is needed to know about nanotubes for understanding the EFM measurements in the upcoming sections. Before presenting the results, however, the methods used for preparing the samples and carrying out the experiments are described. 2 Thicker carbon nanotubes are, however, available in the form of multi-wall carbon nanotubes (MWNT’s) where the thick shells are stabilized by thinner tubes formed in their interior. 83 Gas Sources Tube Furnace Control unit Pressure Gauges Mass-flow controllers Figure 3.3: (a) The CVD oven. (b) picture of ceramic boat with wafers 3.2 Experimental Details This section is devoted to describing the preparation of the SWNT samples in suitable form for EFM measurements, i.e., growth of SWNT’s and processing of metal alignment marks to enable easy navigation. Subsequently, the details of the EFM measurements are presented. Sample preparation As described in section 3.1, the geometry with SWNT’s supported on a conducting substrate capped with an insulating layer is the geometry-of-choice for electrical measurements and, therefore, a number of fabrication techniques have been developed to realize it. First of all, the method depends on the form of the tube material, which in turn depends on the method of synthesis. SWNT’s grown from the carbon-vapor produced either in an electrical arc-discharge between two graphitic electrodes or in the vapor produced by laser-ablating a graphitic target, usually come in bulk quantities and are commercially available from various sources. The tubes grown in this way, however, come in the form of thick bundles (Fig. 3.2(d)) and separating them into individual tubes is a difficult task which involves suspending them in an organic solvent (usually di-chloro-ethane) by ultra-sonication. Subsequently, the tubes are deposited on a suitable substrate by allowing a small amount of the suspension to evaporate on the substrate. The first electrical measurements [21, 22, 23, 24] were made in this way with relatively high quality laser-ablation material (Smalley, Rice University). The sonication may, however, damage the tube structure and nowadays most research groups, including ours, primarily use tubes grown directly on the substrate using chemical vapor deposition (CVD). It should be noted, though, that the various EFM-characterizations demonstrated in this chapter apply equally well to samples prepared from bulk SWNT material. For CVD-growth of SWNT’s the substrates are initially prepared with catalyst particles. A wafer of highly doped silicon capped with 500 nm of SiO2 84 is cut into handleable sizes (∼ 7 × 10 mm, Fig. 3.3(b)) and catalyst particles of iron nitrate (Fe(NO3 )3 ) are deposited following Ref. [25]: Iron nitrate is dispersed in IPA (2-propanol) (∼ 1 mg in 20 ml IPA - solution appears faintly yellow) and each wafer is dipped in the solution for a few seconds followed by a 10 sec. dip in n-hexane and left to dry in air. The dippingprocedure can be repeated for a denser catalyst deposition. Figure 3.3(a) shows the CVD-setup used for growing SWNT’s. A commercial tube furnace3 is equipped with three separate gas lines - one for Argon (Ar), which acts as an inert carrier/diluter of Methane (CH4 ), acting as the carbon feedstock and Hydrogen (H2 ) which reduces the catalyst Fe particles. The flow rate for each gas is controlled by mass-flow controllers4 and after loading the prepared substrates in the oven, a typical growth-run involves: 1) heating of the furnace to 950 ◦ C under Argon, 2) 10 min. of H2 -flow (0.25 l/min), 3) 5 min CH4 (0.75 l/min) and 4) cool down under Argon. Usually the cooldown time was shortened by external air cooling. It has been difficult to get complete reproducibility in the growth but it seems that the growth is selfterminating; i.e. extending the growth-time (step 3) does not increase the density of the growth product. We speculate that this is due to oxidation of the catalyst particles stoping the growth and, indeed, hydrogen reactivates the growth (removes the oxide); i.e., repeating the steps 2) and 3) usually results in denser growths5 After the growth, the samples can be processed using the standard techniques described in section 2.3, and for many of the EFM measurements described in this chapter, a grid of metal alignment marks were fabricated by e-beam lithography, metal evaporation (usually 10 nm Cr, 40 nm Au) and subsequent liftoff. For measurement using the scanning probe microscopy such alignment marks are essential for navigating on the sample and for relocating areas of interest after changing the cantilever. The scanning probe microscope Figure 3.4(a) shows an image of the scanning probe microscope6 (SPM) used in all the measurements presented in this chapter. In addition to the standard features of an atomic force microscope (AFM), the instrument is equipped with an automatic xy-sample stage, camera and for vibrational damping the microscope is supported on a heavy table floating on pressurized pistons and everything is acoustically shielded during operation. The 3 Carbolite MTF tube furnace Mass-flow controller:Brooks Digital Thermal Mass Flow Meter & controllers 5850S. Electronics: Brooks Gas and Liquid Mass Flow Secondary Electronics, Model 0154 5 Thus, mixing the methane with hydrogen during growth may result in longer tubes, however, it has not been possible to reach any conclusions. At present, repeating the steps 2) and 3) seems like the best way of increasing the tube density. 6 Digital Instruments/Veeco Metrology Group, Dimension 3100 Scanning Probe Microscope 4 85 Control electronics (a) (b) 225 µm Acoustic Shield AFM 10-15 µm (b) Vibrationally damped support Figure 3.4: (a) Image of the microscope and controllers. (b) SEM image of the conducting cantilevers used in the EFM studies (SCM-PIT, images from www.veecoprobes.com). AFM is one of the fundamental experimental tools of nano-physics and the following assumes a basic knowledge about its operation. For EFM measurements, a conducting cantilever is required and Fig. 3.4(b) shows an SEM image of those used in these studies7 . For high-resolution topographic measurements, standard tapping mode cantilevers8 tend to give better results (i.e. less noisy) and these have occasionally been used as well. The procedure for tuning the resonant frequency of the cantilever and achieving good performance of the microscope in usual tapping mode follows the standard operation of an AFM and will not be described here; typical parameters can be found in appendix E. For acquiring EFM data, the microscope software’s9 interleave-mode is activated and the desired lift height is entered in the associated field. This enables the dual-scan mode as discussed in below and the desired parameters for the interleave scan can be chosen (tip/substrate voltages, lift-height, scan sizes and velocities etc.). In order to study the EFM response of the same CNT while varying the parameters, the slow scan axis is disabled; i.e., the same line is scanned continuously. To study, for example, the dependence on the tip bias Vt , usually a sequence of 10 scans for each value of Vt followed by a few scans with Vt = 0 V was used. The Vt = 0 V scans make it easy, in further analysis, to distinguish the data-blocks belonging to different values of Vt . A typical image resulting from this procedure is shown in Fig. 3.5(b). The microscope control software 7 SCM-PIT cantilevers from Veeco Instruments, www.veeco.com. Spring constant, k ∼ 2.8N/m, Resonant frequency ω0 ∼ 70 kHz. 8 RTESP, MPP-11100-10, k = 40N/m, ω0 ∼ 270 kHz. 9 Nanoscope V6.11r1, Digital Instrument/Veeco 2003 86 (a) (b) Vt = 8V Vt = 0V Vt = 8V Vt = 6V Vt = 0V Vt = 4V Vt = 0V 1 µm Φ [deg.] Φ [deg.] -4 0 ... 0 -4 Figure 3.5: Illustration of the method used when investigating the dependence of the EFM signal on various parameters. (a) Typical EFM image of CVD grown carbon nanotubes (see section 3.3). (b) Measuring along the dashed line in (a) with the slow scan axis disabled. The tip voltage Vt is changed allowing for ∼ 10 scan lines for each voltage and leaving ∼ 2 scan lines with Vt = 0V to separate the blocks of data. as well as other SPM software packages10 is designed to display the data as images and provides a number of image-processing tools. However, for statistical data analysis of the individual scan lines, as called for by the data in Fig. 3.5, the tools provided by these programs are very limited. For further analysis the data were usually exported to software capable of handling large matrices. 3.3 The EFM technique In this section we consider the principle of the EFM technique. Most of the results derived here can also be found in Refs. [8, 7]. Figure 3.6(a) shows a schematic illustration of the EFM operation of an atomic force microscope. For each scan line, the topography is first attained using standard tapping-mode AFM. The line is then retraced while using the topographic data to keep a fixed tip-surface distance h. During the second scan the tip is oscillated at its free resonance frequency ω0 and the EFM signal is then the measured phase difference between the driving force and the actual oscillation of the cantilever. Letting γ denote the damping coefficient of the oscillation it is a well known result of the damped-driven harmonic oscillator[26] that this phase difference is given by tan(φ) = γω/(ω02 − ω 2 ) as shown in Fig. 3.6(b) (solid curve). At resonance, φ = π/2 and it is standard convention in scanning probe measurements to measure phase-shifts Φ with respect to this value. In the presence of a force gradient F 0 , the resonance frequency changes and as illustrated in Fig. 3.6(b) (dashed curves) a phase 10 The free program WSxM from Nanotec Electronica (www.nanotec.es) has often been used for imaging processing. 87 (a) (b) ω0 π Vt h Φ φ SiO2 Φ t 0 CNT p++ Si ω0 ω Figure 3.6: (a) Side-view schematic of the EFM operation of an atomic force microscope (see text). The black shape illustrates the AFM cantilever and the sample geometry is that of Fig. 3.1(b). difference develops. It is straightforward to show that this phase difference (in radians) is given by Q (3.1) Φ ≈ F 0, k where k and Q = ω0 /γ are the spring constant and quality factor of the cantilever, respectively. For the EFM operation, the SPM is equipped with a conducting tip and letting C denote the tip-sample capacitance, and Vt an applied tip-sample bias, we get F 0 = 21 C 00 Vt2 and thus the EFM signal becomes Q 00 2 C Vt (3.2) Φ= 2k This is an important relation for the analysis of EFM data as it distinguishes between capacitive forces (∝ Vt2 as above) and the force on the tip due to static charges on the substrate (∝ Vt , see section 3.5). The conventions used in SPM for the sign of the phase shift is such that for the attractive force between the cantilever and a conducting sample the phase shift appears negative[8]. Topographic imaging of carbon nanotubes requires high resolution and relatively slow scanning since the typical SWNT feature size is on the order of the diameter ∼ 1 nm close to the typical noise level (usually a few Å) and the roughness of typical substrates used for nanotube devices (SiO2 ). Electrically, however, their conductivities are much larger than that of the insulating substrate (see below) and the tip-sample capacitance is greatly affected by the presence of a nanotube on the surface. Thus, through Eq. 3.2 this translates into contrast for the EFM image. This is apparent in Fig. 3.7(a) and (b) showing a topographic and corresponding EFM image of a sample of CVD grown SWNT’s. The observed branching and the apparent height dt ∼ 1.5 nm (Fig. 3.7(c)) show that the object is probably a thin bundle containing a few tubes. As expected, the bundle appears in the EFM image with a pronounced contrast and due to the longer range of 88 (b) 5 0 z [nm] Φ [deg.] (a) 1 µm (c) (d) Φ [deg.] 0 -2 3 -3 9 -4 6 0 Vt [V] 8 x [µm] 5 10 W Φ0 -1 (f) FWHM [µm] (e) 3 7 z [nm] Φ [deg.] 0 |Φ0| [deg.] 1 µm -4 0 2 1 6 7 0 Vt [V] 8 x [µm] 9 5 10 100 200 h [nm] 300 0.6 0.4 0.2 0 0 100 200 h [nm] 0 300 Figure 3.7: (a) Topographic and (b) EFM of SWNT’s grown on SiO2 by chemical vapor deposition. (c) Lower curve shows the hight profile (right axis) along the black line in (a) across a ∼ 1.5 nm SWNT bundle. Upper curve shows corresponding EFM data (dashed line in (b)) for Vt = 6 V: The red dots show data from 10 scans for h = 30 nm and the black curves show the average for h = 30, 40, 50, 75, 100, 125, 150, 200, 250, 300 nm (bottomtop). (d) The average EFM signal for h = 75 nm, Vt = 6 V (black) and a Lorentzian fit (red line). The definition of the EFM amplitude Φ0 and FWHM W of the CNT-signal is indicated. (e) The fit amplitude W and (f) width Φ0 as a function of lift height for Vt = 6 V (2) and of Vt for h = 100 nm (◦). In (e) solid lines show quadratic fits to Φ0 (Vt ) and Φ0 (h) ∝ h−1 and in (f) the solid lines show linear fits to the signal widths. 89 the Coulomb interactions the bundle appears with a large apparent width W ∼ 400 nm. As a direct consequence, EFM images can be obtained much faster and require lower resolution making the EFM technique advantageous for larger scale sample characterization (see below). The line shape of the EFM response depends on the exact geometry of the tip[27, 28], however, the data agree well with the Lorentzian fit in Fig. 3.7(e) and this is used for analyzing the results. Figure 3.7(d) shows the amplitude Φ0 of the CNT EFM signal as a function of cantilever voltage Vt and the quadratic dependence of Eq. 3.2 is readily observed. Obviously, the parameters of the fit provide information about the capacitance, but since this section is only meant for introducing basic concepts we postpone such detailed analysis for section 3.5 where the result is needed for analyzing additional data. The height dependence of the EFM amplitude enters in Eq. 3.2 through ∂ 2 C(h)/∂h2 and again depends on the exact geometry of the sample and cantilever. In Refs. [28, 27] the relation Φ0 ∝ h−1 was used to fit the height dependence of Φ0 , however, we have found a considerably better agreement using Φ0 = A(h + B)−3 which is seen in Fig. 3.7(e) to fit the data very well in the range considered here (30 nm < h < 300 nm). As seen in Fig. 3.7(f), the width of the EFM signal increases linearly with lift height and is approximately independent of Vt . Thus, in cases where large lift heights are required; e.g., for the purpose of characterization of CNT’s embedded in thick dielectrics (see section 3.6), the weaker signal can be compensated for by increasing Vt effectively without the loss of resolution. After having investigated the lift height and tip voltage dependence of the EFM signal of a particular nanotube we now consider how the signal depends on the nanotube length l and one dimensional conductivity G0 . Consider the schematic in Fig. 3.8(a). If we let C0 denote the tube-backlane capacitance per unit length we have Ctb = lC0 and considering the capacitive voltage divider we get lC0 Vt Vtt = lC0 + Ctt for the tip-tube voltage Vtt . Thus the EFM signal associated with a SWNT of length l becomes Q 00 ³ lC0 ´2 2 Φ(l) = C Vt 2k tt lC0 + Ctt and putting a = (Vt2 Ctt00 Q/2k)−1/2 we get ³ Ctt −1 ´ Φ−1/2 = a 1 + l , C0 (3.3) i.e. Φ−1/2 depends linearly on l−1 . This relation is verified experimentally in Fig. 3.3(b) showing measurements from 30 well-isolated CNT’s. The noiselevel is indicated on the figure and by extrapolation, the shortest observable 90 (a) Vt (b) 3 ~ 250nm Ctt -1/2 1 -1/2 [deg. Ctb ] 2 Φ0 1µm 0 2 4 6 [µm-1] Figure 3.8: (a) Illustration of the terminology used for analyzing the length dependence of the EFM signal. (b) The dependence of Φ0 on nanotube −1/2 length l for 30 well isolated CNT’s (shown as Φ0 vs. l−1 , see text). ◦ The grey area illustrates the noise level (0.15 ) in the measurement and the shortest detectable tube is estimated to be ∼ 250 nm consistent with the fact that no EFM signal could be distinguished for a l ≈ 160 nm tube. The inset shows a typical EFM measurement. tube for the given experimental conditions is lmin ∼ 250 nm (see section 3.4). In section 3.4 we study how defects in individual SWNT’s can effectively separate long tubes into shorter sections thus causing deviations from Eq. 3.3. In the above considerations, it was assumed that the SWNT’s could be safely regarded as metallic objects. We now show (along the lines of Ref. [7]) why this is a good approximation. If the EFM signal Φ of a SWNT is to achieve its full value, the time, τscan , spend by the scanning cantilever in the vicinity of the tube must be larger than the time it takes for the tube to be fully polarized which is the characteristic RC time τRC of the tube. If R0 is the one-dimensional resistivity of the tube τRC = R0 C0 l2 and thus τscan > R0 C0 l2 (3.4) for a fully observable CNT. As an order of magnitude estimate we treat the CNT’s as infinite conducting wires of radius r ∼ 1 nm situated a distance t = 500 nm above a conducting plane. Then C0 ∼ 2π²0 / log(2t/²r r) ∼ 1 × 10−17 F/µm where ²r = 3.9 is the dielectrical constant of SiO2 . Taking vscan ∼ 25 µm/s as a typical scan velocity and 250 nm as a typical length scale for the tip-tube interaction we find τscan = 1 × 10−2 s. Using these values, and working with the one-dimensional conductivity G0 rather than R0 , Fig. 3.9(b) shows the limits set by Eq. 3.4 as well as lmin of Fig. 3.9(a). The shaded area corresponds to combinations of (G0 , l) which are fully ob91 10 Observable CNT's 4 Metallic l [µm] 6 Semiconducting 8 2 0 10-22 10-19 10-16 10-13 G0 [cm/Ω] 10-10 10-7 Figure 3.9: Illustration of the limit on G0 = 1/R0 for a CNT of length l set by eq. 3.4 and lmin of panel Fig. 3.8. The shaded area corresponds to pairs of (G0 , l) for fully observable CNT’s. Also indicated is typical conductivities of (defect-free) metallic and semiconducting CNT’s. servable with the EFM technique. The typical conductivities of metallic (10−9 cm/Ω) and semiconducting (10−11 cm/Ω) SWNT’s[29] are also indicated and the figure clearly shows that both types are expected to be fully observable in an EFM measurement. As mentioned above one of the major advantages of EFM for research involving carbon nanotubes is the possibility of imaging and characterizing CNT samples much faster than with conventional AFM, and in the following sections we investigate what information the EFM technique can provide. We conclude this section by showing in Fig. 3.10 a ∼ 250µm × 330µm EFM image of a CVD grown CNT-sample. The image was automatically acquired in about ∼ 12 hours and consists of 12 separate 90µ×90µm images subsequently stitched together. Individual CNT’s can clearly be resolved and the image provide information about the density, length, orientation of the CNT’s as well as the homogeneity of the distribution - information which would be exceedingly time consuming to obtain with other techniques. 92 10 µm Figure 3.10: Large EFM image (∼ 250µm×330µm) of a CVD-grown sample of carbon nanotubes. The image was acquired in ∼ 12 hours and demonstrates the ability of EFM to investigate the nanotube distribution on a large scale. 93 3.4 Identifying defects using EFM Due to the unique one-dimensional nature of carbon nanotubes the performances of CNT-based electrical devices are dramatically affected by the presence of structural defects. For instance it has been shown that geometric defects such as kinks can make an otherwise metallic tube show semiconducting behavior[30] and defects induced by an electric discharge from an atomic force microscope can completely destroy the electrical conduction of a tube device[31]. Moreover, defects can make nanotubes increasingly sensitive to their electrostatic environment which can be exploited for better performance of nanotube based chemical sensors[32] and it has been shown that defects improve the field emission properties of multi-wall CNT’s (MWNT)[33, 34, 35]. It is, therefore, important to study defects in CNT’s and for many applications ways of assessing defect-related properties on a single nanotube level prior to device fabrication would be beneficial. Most characterization tools, however, either require electrical contact (such as scanning tunneling microscopy and direct transport measurements), are difficult to interpret (micro Raman spectroscopy), or require freely suspended CNT’s (high resolution transmission electron microscopy11 ) which is an inappropriate geometry for standard device applications which usually involves CNT’s supported on insulating substrates as discussed in the previous section. As was also demonstrated in the previous section, EFM provides a noninvasive technique for assessment of electrostatic properties of individual carbon nanotubes in this geometry, and in this section we demonstrate that EFM can provide useful information about their defects. This ability is rooted in the dependence of the EFM signal on the length (Fig. 3.8) and the conductivity (Fig. 3.9) of the nanotubes. A nanotube containing a defect, say halfway along its length, which effectively separate the tube into two conducting segments, will exhibit an EFM amplitude which does not correspond to the CNT’s full length, and from Fig. 3.8 it is possible to assess the actual length of the conducting segments. If the defect is not situated halfway along the tube (as is usually the case) its position will be directly evident in the EFM image as a discrete jump will occur in the EFM phase Φ upon passing the defect. Below we initially discuss measurements where a mild oxygen plasma etch is used for intentionally inducing defects in nanotubes and we show that these are easily detectable by mapping individual nanotubes using EFM. Subsequently, it is demonstrated that the general principle can be used to assess the quality in as-grown samples where defects are not intentionally induced. 11 Note, that TEM will most likely also damage the tubes in the process of imaging. 94 (a) (b) 2µm (c) Intensity [A.U.] (d) Pristine Etched G D 1600 1400 1200 Frequency [cm-1] Figure 3.11: EFM images of CVD grown CNT’s (a) before and (b) after a brief oxygen plasma etch. The double-square in the center is a metal alignment mark used for navigation on the sample. Most of the CNT’s loose their EFM signal by the ash (dashed lines). No change is found in the topographic data (c). The dark spots seen in the topographic image are minor (5-10 nm) pits in the oxide substrate and are unimportant for this work. (d) Normalized Raman spectra using an excitation energy of 1.17 eV of bulk CNT’s before (red) and after (black) plasma etching. The enhanced D-band of the etched tubes indicates an increase in the defect density (see text). Plasma Induced Defects The technique of oxygen plasma etching12 is widely used for removing organic material and it completely removes the CNT’s from the CVD samples within ∼ 60 − 90 sec. For much shorter times, however, such an etch is expected to leave the bulk of the CNT’s intact and merely introduce occasional defects. Figure 3.11(d) shows Raman spectra measured at four different areas of a CNT13 sample before (red traces) and after (black) a brief (∼ 1 sec.) oxy12 A commercial plasma system Plasma-Preen II-862 from Plasmatic Systems Inc. operating at a O2 pressure of 1-5 torr and a power of ∼ 600W was used for this study. 13 For Raman measurements samples were prepared by deposition of HiPCO CNT’s[36] (rather than CVD-growth) in order to achieve higher nanotube densities and thereby increase the intensity of the Raman signal. Substrates for the Raman measurement were 95 gen plasma etch14 . The spectra have been normalized with respect to their G-band intensity and the (near) indistinguishability of the individual spectra in each group (emphasized in the inset) shows the homogeneity of the distribution of tubes probed by the measurement. The double-peak structure of the G-band shows that the sample consists of single-wall CNT’s and since the processes of the D-band are mediated by scattering events involving defects[14] the large increase in D-band intensity after the etch directly confirms that this treatment increases the density of defects. Turning to the effect on the EFM-signal, metal alignment marks were fabricated on a sample to allow easy navigation and Fig. 3.11(a) and 3.11(b) show EFM images before and after the etch, respectively. The sample has a relative high density of tubes as is clearly seen in 3.11(a) and the observed branching indicates that they come in form of thin bundles. There is no apparent effect of the etch observed in the topographic data (Fig. 3.11(c)), however, in panel 3.11(b) it is clearly seen that it causes the EFM-signals of most of the nanotubes to drop below the noise level. There is no change upon reversing the tip polarity, and repeating the scan after one month yields the same result. This rules out charging effects as the cause of the loss of EFM signal[6] and leaves defect-induced changes in the CNT properties as a natural explanation. The signal remains only for the thickest bundles probably because the outermost tubes of the bundle partly protect the inner ones from the oxygen plasma. Notice that in some cases the signal disappears abruptly, and no tubes with homogeneous but faint signals are observed. This indicates that for the EFM signal, the most significant effect of the etch is to introduce a number of discrete defects which effectively separate the tubes into conducting sections shorter than lmin ∼ 250 nm rather than inducing a homogeneous defect distribution which would increase the scattering and suppress the overall conductivity below Gmin . This is consistent with the study of Ref. [31] where it was shown that defects not observable in topographic data could, nevertheless, reduce the conductance of a CNT to zero. These results demonstrate that EFM, in some situations, can reveal defects in CNT’s which would pass unnoticed in standard topographic measurements. We note that our findings of the etch making the tubes highly insulating, may add to the explanation of the improved field emission properties of plasma etched MWNT’s[33, 34, 35]: Etching MWNT’s will affect the outer walls first thus leaving a thinner (and thereby more strongly emitting) conducting core to act as the emission source. covered by an evaporated layer of Au to prevent an interfering signal form the Si and amorphous SiO2 layer. HiPCO tubes were ultrasonically suspended in di-chloro-ethane and a small amout left to dry on the substrate. The tubes precipitate as small grains on the surface which can be seen in the Raman microscope. 14 The plasma is turned off immediately after it is lit making the effective etching time as short at possible (. 1 sec.) 96 (a) (b) 4 4 5 6 5 6 3 3 1 2 1 2 1µm Φ0 [deg.] (c) 4 4 2 2 5 3 1 0 1 x [µm] 4 Figure 3.12: Topography (a) and corresponding EFM image (b) of an asgrown CNT sample with defect-related effects indicated by arrows (see text). (c) EFM signal Φ0 along the CNT indicated by arrows 1-5 in (a): The section 2-3 has no EFM response and at 4 a defect causes an abrupt change in Φ0 . Defects in as-grown samples In Fig. 3.12 we explore applications of EFM in situations where defects are not intentionally induced. Figures 3.12(a) and (b) show topography and EFM image, respectively, from an as-grown sample of CNT’s and by carefully correlating the two images several defect related features are observed. Consider first the CNT indicated by arrows 1-5. The section 2-3 is significantly thinner than the rest and as seen in 3.12(b) it is completely absent in the EFM signal showing that this section is probably highly defective or poorly graphitized. Without the EFM data, however, one might wrongly attribute the structure observed in the topography to a rope which along 2-3 just contains fewer tubes. The CNT section 3-5 shows no obvious defectrelated features in the topographic image. In the EFM image, however, an abrupt intensity increase occurs at 4, as is emphasized in panel (c) showing Φ0 (x) along 1-5. Thus, the EFM data show that at the point 4 the tube contains a defect which separates the segment 3 − 4 from the rest and since this segment is shorter (l3−4 ≈ 1.4 µm) than the remaining part of the tube it appears with a weaker EFM signal. The combination of length and EFM signal of the segment 3 − 4 agrees with the linear relation in Fig. 3.8(b) showing that the defect at 4 is highly resistive (i.e. transport across the defect at 4 occurs with conductivity ≤ Gmin ) and that the section 3 − 4 is by itself not separated into even shorter sections. For more examples of defects in as-grown samples and the effects on the EFM signal see Ref. [12]. We note, that the techniques described here could be used for assessing the 97 frequency of defects in nanotube samples grown under various conditions information which would aid in optimizing the synthesis of SWNT’s. 98 3.5 Charge trapping in Carbon Nanotube Loops demonstrated by EFM The electronic interaction between carbon nanotubes and local charges plays a vital role for applications of CNT’s in molecular electronics. Carbon nanotube field effect transistors (FET’s) exhibit hysteric behavior upon sweeping the back gate due to rearrangements of charges or ions in water which locally gate the semiconducting nanotube[37, 38, 39]. This hysteresis can be exploited in CNT-based memory elements[37, 38] but must be eliminated for reproducible operation of the CNT FET’s in electronic circuitry. Also applications of carbon nanotubes as the active element in nanoscale sensors for chemical compounds[40, 41] exploit the high sensitivity of carbon nanotube FET’s to local charges in the environment[42]. In the previously reported studies of these effects, the role of the charges has been elucidated by monitoring the change in conductance of CNT FET’s upon changing the nanotube environment. In this section we demonstrate the use of EFM to investigate the interaction of surface charges on SiO2 with carbon nanotubes. We find that tubes strongly influence the dynamics of surface charges and that charges can be trapped on regions of the substrate fenced by nanotubes. Figure 3.13(a),(b),(c) show EFM-images of SWNT’s grown by CVD as described in section 3.2: For this particular growth, the CNT’s are extremely long (up to ∼ 100 µm) and form loops and coils, as also reported in Ref. [43] for long CVD grown CNT’s. An example of a nanotube coil formed at the end of a CNT is shown in Fig. 3.13(b). As seen in Figure 3.13(a),(b), a pronounced EFM signal Φl is observed in the interior of some of the loops. This reflects that the tip experiences a force gradient over the substrate enclosed by a nanotube loop different from when it is outside a loop. This force cannot be attributed to a mere capacitive coupling to the nanotubes since it is often observed that in two loops made from the same tube and of similar geometries only one exhibits the contrast in the interior. Such cases are marked by the arrows in Figure 3.13(a). Figure 3.14(a) shows the phase shift along a line through a nanotube coil with radius r ∼ 2µm (Figure 3.14(b)). Sharp peaks ΦN T are observed when the tip passes by a nanotube, and the enhanced phase difference Φl in the interior of the coil is clearly observed. Figure 3.14(c) shows how Φl and ΦN T depend on the potential difference Vs between tip and substrate15 . The relation ΦN T ∝ Vs2 derived in section 3.1 (equation 3.2) and expected for a capacitive coupling to the carbon nanotubes is clearly observed, but for the interior of the loop we find Φl ∝ Vs , indicating different physical origins for the two forces. We ascribe Φl to static charges trapped in the interior of 15 Note that for the measurements in this section the voltage was applied to the substrate in interleave mode while keeping the tip grounded rather than the reverse as indicated in Fig. 3.6. This has no practical significance, however, in this section we work with Vs rather than Vt . 99 (a) 50 z [nm] 100 150 (d) 200 Phase shift [deg.] 0 -5 -10 -15 -20 -25 0 20 40 60 80 100 2 2 Vs [V ] (c) (b) Figure 3.13: (a) EFM image of long CVD-grown nanotubes forming loops. Enhanced EFM phase-shift Φl is observed from the interior of some of the loops. Arrows point to pairs of similar loops formed on the same CNT of which only one exhibit enhanced Φl . (b) EFM-image of CNT loop and coil from a different spot on the sample. (c) EFM image showing many loops. (e) Phase difference over the bare substrate as a function of Vs2 for z = 60nm (¤) and lift height z for Vs = −10V (•). The data has been fitted to theory, see text. Scale bars for all images are 10µm, intermediate lift heights are z = 60nm, and Vs = −6V . the nanotube loops. If a charge Qs rests on the substrate, image charges qt = AQs and qb = (1 − A)Qs will be induced in the tip and back gate, respectively. The parameter A ∈ [0, 1] describes the division of the image charges between tip and substrate. With a capacitance Ctb between tip and backgate the total charge on the tip is Qt = qt + Ctb Vs and the force on the tip can be approximated by[44, 45] F (z) ≈ −Qs (AQs + Ctb Vs ) 1 0 2 − Ctb Vs . 4π²0 z 2 2 (3.5) Here we have neglected the van der Waals forces which are negligible for z > 10nm [46]. The force gradient entering equation 3.1 is then · ¸ Qs Vs Ctb 1 ∂Ctb AQ2s 1 00 2 0 F = − + − Ctb Vs . (3.6) 2 3 2π²0 z z 2 ∂z 2π²0 z 2 The last term is canceled by the offset of our measurement. Thus, for the static charge contribution, F 0 depends linearly on Vs as we measure inside the loop. To estimate the charge density in the loop we must determine the capaci100 (a) (b) (e) Phase shift [deg.] 0.2 Φl 0,0 -0.2 ΦNT -0.4 -0.6 (f) -0.8 0 2 4 6 8 10 12 14 16 18 x [µm] 0.0 (c) (d) Φl 0.0 Phase shift [deg.] Phase shift [deg.] 0.5 -0.5 -1.0 ΦNT -0.5 -1.5 -6 -4 -2 0 2 4 6 8 -1.0 10 Sample bias [V] 40 80 120 160 200 Lift height [nm] Figure 3.14: (a) EFM phase shift with Vs = −5 V through a nanotube loop along the line in (b) for Vs = −5 V. x = 0 µm corresponds to the top of the line. The two edges of the loop is seen as sharp peaks with phase shift ΦN T and the trapped charges give rise to shift Φl from the interior of the loop. (c) The phase shifts Φl and ΦN T from (a) as a function of sample bias. The measured points are averages from ∼ 15 scans along the line shown in (b). A linear fit to Φl and a quadratic fit to ΦN T are shown as solid lines. (d) Phase shifts Φl for the loop shown in (e) and (f) [AFM and EFM, respectively] as a function of lift height for Vs = −5V (¥) and VS = −10V (¤). The solid lines are fits to powers az −3 + b. Scale bars in (b) and (e) are 5 µm and (b), (f) are measured with Vs = −5 V and z = 60 nm. tances called for in equation 3.6. Figure 3.13(c) shows Φ vs. Vs measured over the naked substrate. Again, the Vs2 dependence is found and from the slope of the linear fit (2.8 × 10−3 rad./V2 ) and equation 3.2 we find 7.0 × 10−5 F/m2 for the second derivative. This, however, corresponds to the entire cantilever-substrate capacitance, but at these heights only about one third of this is due to the tip of the cantilever which is the part sensitive to the charge[47]. Thus, approximating the tip as a circular disc of radius R and the tip-substrate system as a parallel plate capacitor with z = 60 nm air and t = 400 nm SiO2 with dielectric constant ²s = 3.9²0 , we estimate[8] 00 Ctb = 2πR2 ²0 F = 2.3 × 10−5 2 , 3 (z + t/²s ) m (3.7) 0 = − 1 C 00 (z + t/² ) = −1.9 × 10−12 F/m and and from this we find Ctb s 2 tb 0 (z + t/² ) = 3.1 × 10−19 F. In this approximation, the tip acts as Ctb = −Ctb s 101 a circular disc with effective radius r ≈ 40 nm in reasonable agreement with the observation that CNT’s appear in EFM-measurements with widths of about 200 nm which is close to 2R (Fig. 3.14(a)). In Fig. 3.13(c) we also plot the z-dependence, and fitting the data to equation 3.7 (solid line) we find again effectively R ≈ 40nm identical to the estimate from the voltage dependence. Using these capacitances in equation 3.6 and relating the measured phase shifts to F 0 by equation 3.1 we use the slope of the line in Fig. 3.14(c) to calculate Qs ≈ −7.7 × 10−19 C. The sign of the charge is inferred by noting that conducting objects (e.g. CNT’s) which attract the tip appear with a negative phase shift, and negative Φl is measured with Vs negative, i.e. with a positive charge on the tip. Thus, the loop charge Qs must be negative. This charge is situated within an area πR2 which gives a charge density of 2.2 × 10−8 C/cm2 , i.e., the coil on Fig. 3.14 contains a charge of about 17500 e− . With a similar analysis it is found that the r ∼ 1µm loops in Fig. 3.13(a) contain between 1500 and 3500 e− each, corresponding to densities around 0.8 − 1.8 × 10−8 C/cm2 . Studies of charges induced on sapphire substrates by corona discharge from an SPM-tip[48] yield densities of 2 × 10−8 C/cm2 close to our measured value. The third term of equation 3.6 gives the Vs -independent phase shift due to the charges on the surface. Using our calculated Qs , we find a value of 0.1×A deg. which is consistent with the fact that we do not observe a phase shift for Vs = 0 V if 0 < A < 0.1. Figure 3.14(d) shows the height dependence of Φl for Vs = −5 V and Vs = −10 V for the loop shown in Figure 3.14(e),(f). Assuming the Coulomb interaction dominates the h-dependence we have fitted the data to the function ah−3 + b, i.e. the first term16 of equation 3.6. The fit gives charge densities of 9.9 × 10−10 C/cm2 and 1.8 × 10−9 C/cm2 for the Vs = −5 V and Vs = −10 V respectively. Using the Vs -dependence as above we find a charge density in the loop of ∼ 2.4 × 10−8 C/cm2 , i.e., an order of magnitude larger. We expect the estimate from the h-dependence to be the less accurate since the relative scales of the tip-sample-substrate system changes when altering h and this will influence the interaction of the charges with the tip. Further evidence that Φl is indeed due to charges on the substrate is provided by AFM manipulation. Figure 3.15(c) shows an AFM image of two loops formed on the same nanotube which initially each hold ∼ 6500 e− trapped in their interior as seen on the EFM-image (panel a). By touching the substrate in the interior of the lower loop with the grounded AFM-tip the loop is discharged as observed on the subsequent EFM-image (panel (b)). It was not possible to recharge the loop by applying voltages of ± 20V to the tip while touching the substrate. The disappearance of the loop force unambiguously shows that Φl is not due to a geometric effect. Furthermore, 16 The fit remains the same if the entire expression of equation 3.6 is used. 102 (a) (b) (c) (d) (e) (f) (g) Figure 3.15: (a) EFM image of two loops formed on the same nanotube, each trapping charges as seen by the phase difference measured in their interior. (b) EFM-image of same area as (a) after touching the interior of the lower loop by the grounded AFM-tip. (c) Topographic image of the nanotube showed in (a) and (b). (d)-(f) AFM and EFM images of nanotube coil before (d),(f) and after (e),(g) the coil is broken by AFM manipulation. All measurements are performed with z = 60nm and Vs = −6V . Scale bars are 2µm. since the entire loop is discharged by just touching the middle of the loop and the observation that the charges are uniformly distributed within the loops suggests that the charges are relatively mobile on the surface and not trapped deeply in the substrate. In panels (d)-(g) a similar series of images is shown where the charges (∼ 3000 e− ) are removed by cutting a nanotube coil using the AFM-tip in contact mode to ensure that the surface was indeed touched when removing the charges. Only a fraction of all loops have trapped charges as seen from Fig. 3.13(d) which shows 22 loops on a different sample, none of which show trapped charges, and Fig. 3.13(a) which shows loops both with and without charges. We have not observed significant trapped charges in regions fenced by CNT’s in other geometrical configurations than closed loops or coils. We speculate that the charges stem from ions present during the CVD growth of the nanotubes. The CNT may deplete neighboring regions of the substrate from free charges, effectively making the region in the interior an isolated island where charges can reside. Alternatively, the CNT loops could themselves be 103 negatively charged and thus trap charges through the Coulomb interaction. In the latter case, however, it should be possible to observe by EFM the static charges on the CNT which has not yet been possible. It it also possible that the hydrophobic CNT encloses a thin layer of adsorbed water molecules on a surface which can contain ions [39]. Tribological effects where charges are induced on the substrate by the scanning cantilever can be ruled out since scanning does not recharge loops which were once discharged as in Fig. 3.15(a). In the lower loop of Fig. 3.15(a) charges did not reappear after 12 hours. Further work where charge is deliberately injected during the experiments could aid in resolving the issue. In conclusion, we have used EFM to quantitatively address the interaction between surface charges and carbon nanotubes. We find that surface charges can be effectively trapped on the substrate in regions enclosed by CNT’s. Static charges are responsible for the hysteretic behavior of CNT FET’s which play a key role in CNT memories, sensors and logic circuits. This suggests that the hysteretic behavior of CNT-loop FET’s might be different from that of ordinary CNT FET’s. Furthermore, we have shown that EFM provides a powerful tool for rapid characterization of large area CNT samples and the present work shows how to identify contributions from static charges in such data. 104 3.6 EFM for assessment of embedded CNT’s Individual single wall carbon nanotubes posses extraordinary mechanical, electrical and thermal properties and great effort has been put into transferring these properties to bulk systems by incorporating SWNT’s in polymer matrixes[49, 50]. Due to difficulties in solubilization and chemical functionalization of SWNT’s, as well as strong tendency for bundle formation, it has proven difficult to produce high quality suspensions of SWNT’s in polymers. Furthermore, it is difficult to evaluate, on a single nanotube level, the structure of such polymer suspensions since the polymer matrix makes the use of conventional characterization techniques questionable. Atomic force microscopy (AFM) and scanning electron microscopy (SEM) are capable of imaging only tubes at, or very close to the surface[51, 52, 53] and transmission electron microscopy (TEM) requires delicate sample preparation[49]. As is apparent from the first section of this chapter, EFM is capable of assessing nanotubes imbedded in an insulating matrix since the technique relies on the electrostatic interaction. In this section we demonstrate the use of EFM as a simple, non-invasive tool for mapping the three dimensional orientation and position of individual SWNT’s in thin films of SWNT/polymer composites. The technique is used for studying spin-cast films of the composite formed by ultrasonically suspending SWNT’s in poly-(methyl methacrylate) (PMMA)[54, 53]. It is demonstrated that EFM technique reliably provides statistics on the density as well as the length and orientation distribution of the suspended nanotubes. Towards the end we discuss the limitations of the technique and it is shown that for long tubes & 3 µm composite samples with thicknesses up to ∼ 1 µm can be imaged. Basic characterization of composites The samples studied here differ from the samples of the previous sections and were prepared in the following way. A small amount (< 1 mg) of commercially available SWNT’s17 grown by the electrical arc technique was suspended in a 2% solution of poly-(methyl methacrylate) (PMMA) in anisole by low power ultrasonication for 30 minutes. The resulting homogeneous black suspension was immediately spin-casted on wafers of highly doped silicon capped with 500 nm of insulating SiO2 by spinning at 3000 rpm for 45 sec. and subsequent evaporation of the anisole on a hotplate (185 ◦ C) for 90 sec., This results in a 50 − 60 nm film of the SWNT/PMMA composite on the substrate as schematically shown in Fig. 3.16(a). Figure 3.16(b) shows an AFM topography image of the resulting surface. 10 − 20 nm particles are observed probably due to the amorphous carbon ash present in the nanotube material, however, no nanotubes can be identified on the surface. Due 17 As-grown material from Carbon Solutions 105 15 180 #counts (d) 20 #counts 0 15 0 (c) (a) Vt 90 θ [deg.] h PMMA 10 Si/SiO2 0 1 2 3 L [µm] 4 0 10 -4 z [nm] (b) Φ [deg.] 0 0 5 µm 5 µm Figure 3.16: (a) Schematic illustration of the spin-cast polymer/SWNT sample and EFM operation of a SPM (see text). (b) AFM image of the surface topography of a ∼ 60 nm thick film of PMMA/SWNT composite. Due to the polymer the tubes cannot be observed, however, in the corresponding EFM image (h = 35 nm, V = 7 V) in panel (c) individual embedded SWNT’s are clearly seen as dark lines. (d) The distribution of the lengths and orientation (inset) of 95 embedded SWNT’s measured in a 90 µm × 90 µm area of the sample (see text). The angle is measured with respect to the horizontal direction in (c). to the much higher conductances of SWNT’s as compared to the PMMA matrix the tip-sample capacitance is significantly altered in the presence of a SWNT sandwiched between the tip and the substrate. The embedded SWNT’s are therefore clearly revealed in Fig. 3.16(c) showing the EFM image corresponding to the area in (b). Since the individual nanotubes can be identified, important parameters characterizing the nanotube distribution can be reliably obtained. In Fig. 3.16(d) we show the distribution of lengths and in-plane orientations for 95 individual tubes measured in a 90 × 90 µm area of the sample. The length distribution peaks around ∼ 750 nm and the orientations are almost random, however, with a slight tendency for aligning along the ∼ 50 ◦ directions which is most likely an effect of the spin-casting. We note that as the nanotubes are much longer than the thickness of the composite film the projected lengths measured in the EFM image are very close to the actual lengths (see below). One of the obstacles for preparing nanotube suspensions is the strong tendency of the tubes to aggregate and to form bundles. It is not possible in the EFM image to distinguish a single SWNT from a thin bundle but it is possible to evaluate to which degree the tubes are freely suspended or form larger aggregates. In the image in Fig. 3.16(c) most tubes are isolated but occasionally two or more are touching 106 suggesting that additional sonication may yield a better suspension18 . Detailed investigations and optimizations of the parameters of the composite preparation scheme are, however, not the focus of this study. 3D mapping of SWNT’s in composites For fixed Vt the amplitude of the EFM signal from two identical SWNT buried at different depths in the polymer film depends on the backplanetube-tip distances. Due to the large SiO2 thickness (500 nm) the relative difference in backplane-nanotube distances is small and thus the change in the EFM signal is effectively caused by the relatively larger change in tip-tube distance. In the following we neglect the effects of differences in backplanenanotube distances and use the EFM image to map the three-dimensional position of the SWNT’s in the polymer matrix. For this demonstration a sample consisting of a composite-PMMA-composite (60/50/60 nm) trilayer was fabricated by repetition of the spin-casting procedure described above. Figure 3.17(a) shows an EFM image from this sample measured with Vt = 7 V and a tip/sample separation h = 35 nm. Two SWNT’s, T1 and T2 with lengths LT 1 ≈ 1.5 µm and LT 2 ≈ 1.6 µm are clearly seen. T2 clearly has a lower EFM amplitude ΦT0 2 than T1 . In addition to the distance of the nanotubes from the sample surface the EFM signal depends on their lengths as discussed in relation to Fig. 3.8, and since we are interested in the position of the nanotube in the polymer matrix we first compensate for the length dependence as described below. The inset to Fig. 3.17(b) shows Φ0 vs. L for 30 isolated SWNT’s grown on a SiO2 substrate without polymer19 (i.e., h is constant). The solid line shows −1/2 a fit to the theoretical prediction Φ0 ∝ L−1 derived in section 3.1 and the figure shows that the length differences between T1 and T2 can (as a first approximation) be compensated for by a ∼ 10% decrease in ΦT0 2 . Figure 3.17(b) shows the resulting values of Φ0 for T1 and T2 measured as a function of lift height at the positions indicated in Fig. 3.17(a). For each lift height ∼ 10 scans were measured and the average nanotube signal fitted to a Lorentzian line shape Φ(x) = Φ0 /(1 + 4((x − xc )/W )2 ) with amplitude Φ0 and full-width-at-half-max (FWHM) W . The results for T2 are obviously horizontally off-set by an amount δh, i.e., for T2 the tip-surface distance does not correspond to the tip-tube distance. To estimate δh the data is fit to the relation Φ0 = A/(h + B)3 . The solid lines show fits using A = 0.124 deg./µm for both, and we get δh = BT 2 − BT 1 = 32 nm as the 18 In other samples having received less ultrasonication most tubes are found as large aggregates containing hundreds of tubes. 19 The measurement was performed with h = 60 nm and Vt = 5 V and to compensate for the different Vt the signal has been normalized to coincide with ΦT0 1 measured for h = 60 nm 107 3 T2 Φ0 [deg.] 1 µm T1 Φ0 [deg.] (a) T1 2 4 2 0 0 T2 1 2 L [µm] 1 y (b) x 0 -2 -4 δh 0 Φ [deg.] 50 150 h [nm] 250 (c) T1 z [nm] 0 6 5 4 3 2 -40 -80 y [µm] 1 -120 1 2 3 4 X [µm] 5 6 Figure 3.17: (a) EFM image showing two SWNT’s embedded in a ∼ 170 nm thick film of SWNT/PMMA composite. (b) The lift-height dependence of the length-corrected EFM signal of the two tubes in (a). The measurements were performed at the points indicated in (a) - see text. The measured amplitudes have been fit to Φ0 = A/(h + h0 )3 . The inset shows Φ0 vs. nanotube length L for 30 isolated SWNT’s measured with known tip-tube separation h = 60 nm. The red line shows a fit to the theoretical prediction −1/2 Φ0 ∝ l−1 . (c) A projection view of the three dimensional map of the two nanotubes as inferred from the data in (a) and (b). The blue region illustrates the PMMA matrix (see text). 108 effective distance between T2 and T1 . Since T2 is embedded in PMMA with the dielectric constant ²P = 3.2 this corresponds to T2 being situated a distance δz = ²P δh ≈ 100 nm below T1 . The xy-coordinates along each nanotube is found by tracing the minima in Fig. 3.17(a) and for each minimum the relative z-coordinate is found from the particular Φ0 and the fits in Fig. 3.17(b) (and adding δz for T2 ). A projection view of the resulting map is illustrated in Fig. 3.17(c). Since T1 emerges from a catalyst particle which can be seen in the topographic data as in Fig. 3.16(b) we expect T1 to be situated close to the surface of the sample and the blue region in Fig. 3.17(c) illustrates the PMMA matrix. The z-extension of the tubes agrees with the fabrication of the sample from which we expect nanotubes in the lower and upper ∼ 60 nm of the composite but a nanotube-free middle section of ∼ 50 nm. In Refs. [28, 27] the relation Φ0 ∝ h−1 was found to describe the height dependence of Φ0 . For the data presented here as well as data obtained without PMMA (Fig. 3.7), i.e., where h is known, we find Φ0 = A(h + B)−3 to provide the best fit. However, using ΦT0 1 = A/h to estimate A and using this value in fitting ΦT0 2 = A/(h + δh) yields δh = 32 nm in agreement with the above analysis. We note that once the length and h dependencies are known for a given sample and setup, a single image provides sufficient information for a complete three-dimensional mapping of all the SWNT’s, i.e., the above analysis does not need to be repeated for each SWNT in finding its z-position. Limitations of the technique We now discuss the limitations of the technique for mapping SWNT’s in polymer composites. Firstly, the decreasing EFM amplitude for larger tiptube distances limits the thickness of the composite films which can be studied with the technique. In Fig. 3.17(b) the SWNT signal can be clearly observed up to h = 225 nm, however, we have routinely imaged SWNT’s with h up to 300 nm and taking the dielectric constant of the polymer into account we estimate the limiting thickness to be ∼ 1 µm. We have made no attempts to reduce the noise in the commercial AFM setup but we expect that such efforts will increase the limiting thickness further. Furthermore, since the amplitude also decreases for shorter SWNT’s the length Lmin of the shortest observable SWNT increases with the tip-tube distance. Using the parameters of the fits to the L and h dependencies in Fig. 3.17(b) we show by the shaded area in Fig. 3.18(a) the combinations of L and h which are observable by the EFM technique with the boundary given by the noise level in the measurement. Thus, close to the surface the technique detects all SWNT’s longer than ∼ 250 nm, however, from the most distant areas of thicker samples only tubes with L & 3µm will be observed. Secondly, due to the long range of the electrostatic interaction, the spacial resolution of 109 300 0.6 (a) (b) W [µm] h [nm] 250 200 150 0.4 100 50 0.5 1.0 1.5 L [µm] 2.0 2.5 0.2 0 100 h [nm] 200 Figure 3.18: (a) Graph showing the combinations of CNT length L and effective tip-tube separation h which are observable with the EFM technique (shaded area). The graph is calculated using the parameters of the fits in Fig. 3.17(b) and hatched region illustrates the uncertainty (due to the uncertainty in the values of the fits). (b) The FWHM of the Lorentzian fit to the EFM signal of T1 as a function of the lift height. The solid line is a linear fit. the technique is expected to decrease for SWNT’s further away from the sample surface. This is confirmed in Fig. 3.18(b) showing a clear linear increase of the FWHM of the EFM signal from T1 as a function of tip-tube distance[28, 27]. We note that this dependence can be exploited for reaching a better estimate of the z-position of the SWNT’s20 . Thirdly, the technique as presented here is only suited for samples where the length as measured with EFM is close to the actual tube length, i.e., the SWNT’s must be approximately aligned with the substrate plane. This is, however, naturally achieved by spin casting the composite and is therefore not a major constraint. In addition, we suspect that by including also the width dependence shown in Fig. 3.18(b) and supporting the results with model calculations[55] including the effect of larger tube-substrate angles, this constraint may be further relaxed. Finally, at high filling factors of SWNT’s in the polymer matrix the EFM 3D-mapping method as described here becomes unsuitable since it requires the nanotubes to have well-defined lengths, i.e., each nanotube needs to be electrically isolated. The technique may, however, still provide a valuable tool for imaging the suspended tubes closest to surface. To summarize, we have in this section showed the use of electrostatic force microscopy as a non-invasive technique for the three dimensional mapping of individual SWNT’s in polymer/SWNT composites. The technique is shown to provide important information about the homogeneity of the incorporated SWNT’s and since individual nanotubes can be identified the distribution of lengths and orientation can be measured (information otherwise inac20 The measurement of ΦT0 2 (h) in Fig. 2(b) was not performed with a scan direction exactly perpendicular to the tube axis and thus the width cannot be directly compared with that of T1 in Fig. 3.18(a). Extracting the perpendicular FWHM of T2 from Fig. 3.17(a) gives a width of ∼ 330 nm and with h = 35 nm this corresponds to a horizontal offset in Fig. 3.18(a) of ∼ 45 nm in good agreement with δh deduced from Φ0 (h). 110 cessible). Such 3D mapping has not previously been shown for nanotube composites and we expect that the technique could become a valuable tool for the future studies and developments of these materials. 111 3.7 Conclusion of EFM studies In the preceding sections we have seen the use of electrostatic force microscopy for different studies of single wall carbon nanotubes. The work with the EFM technique originated from an attempt to find the fastest and best way of characterizing our CVD-grown nanotube samples. It was quickly realized that EFM was the easiest way of measuring basic parameters such as densities, lengths, orientation, homogeneity etc. and it was used to build a catalog of the various samples grown. The different studies presented in this section appeared when trying to understand various features in the measured images. It was found that the technique also convey information about the quality of the tubes and their interplay with static charges. Moreover, it was shown that EFM, as the only known technique, can be used for the three-dimensional mapping of individual nanotubes in SWNT/polymer composites. I believe that the technique will find many uses as a powerful tool in the nanotube research and will become recognized as a standard characterization tool of nano science. Considering the differences between the subject of this chapter and that of the first chapter on electron transport in InAs nanowires (at least) two questions would be reasonable to raise; how does EFM apply as a characterization tool for the studies of nanowires and what are the differences in the transport characteristics of the two systems? Concerning EFM as a tool for nanowire characterizations it has not found many uses - EFM may be a superior technique for characterizing nanotubes (as advocated for in this chapter). However, since nanowires are considerably larger (∼ 50 times larger diameter) they can be imaged using standard optical microscopy or by SEM21 , thereby rendering scanning probe techniques obsolete. Also, for EFM studies of nanowires, the large diameter of the wires causes the tipbackgate distances to be significantly altered when scanning a wire and this effect makes interpretation of the EFM images more involved than for the nanotubes, where the tip-backgate distance can effectively be considered constant. We note, however, that the EFM technique (or its close analog Kelvin probe microscopy) may provide important information about the potential landscape if imaging contacted nanowire devices[56] and such experiments are planned for the near future. Regarding electric transport in carbon nanotubes, this has been the subject of numerous studies. Carbon nanotubes constitute a truly fascinating system and many remarkable results have been found. Their properties resemble in many ways those of the nanowires as presented in previous chapter; well-defined quantum dots can nowadays be routinely made in carbon nanotubes exhibiting the physical phenomena of Coulomb blockade, Kondo 21 Also, due to the larger scale of the nanowires SEM imaging is not expected to significantly damage the wires as is possible the case for nanotubes (section 3.4). 112 SWNT loop contacts 1 µm Figure 3.19: AFM image of a looping nanotube with lithographically defined metal contacts for future investigations of electron transport. physics and open quantum dots (Fabry Perot interference). In this respect, the subject of transport in nanowires is still in its infancy. The original aim of this project was to study transport in semiconducting nanowires and transport in nanotubes was not pursued systematically. However, it would be interesting to investigate electron transport in the nanotube loop-structures of section 3.5 for the possible interference of the electrons going around the loop and the influence of a magnetic flux through the loop. Devices were fabricated for these investigations as shown in Fig. 3.19, however, due to limited time (and technical issues) their transport properties have not been measured yet but is scheduled for the near future. 113 Bibliography [1] M. Dresselhaus, G. Dresselhaus, P. Avois, and (editors), “Carbon nanotubes: Synthesis, structure, properties, and applications”, Topics in Applied Physics 80 (2001). [2] S. Iijima, “Helical microtubules of graphitic carbon”, Nature 354, 56 (1991). [3] S. Suzuki, K. Kanzaki, Y. Homma, and S. Fukuba, “Low-accelerationvoltage electron irradiation damage in single-walled carbon nanotubes”, Jpn. J. Appl. Phys 43, 1118 (2004). [4] S. Suzuki and Y. Kobayashi, “Conductivity decrease in carbon nanotubes caused by low-acceleration-voltage electron irradiation”, Jpn. J. Appl. Phys 44(46-49), 1498 (2005). [5] A. Vijayaraghavan, K. Kanzaki, S. Suzuki, Y. Kobayashi, H. Inokawa, et al., “Metal-semiconductor transition in single-walled carbon nanotubes induced by low-energy electron irradiation”, Nano Lett. 5(8), 1575 (2005). [6] M. Zdrojek, T. Melin, C. Boyaval, D. Stievenard, B. Jouault, et al., “Charging and emission effects of multiwalled carbon nanotubes probed by electric force microscopy”, Appl. Phys. Lett. 86(21), 213 114 (2005). [7] M. Bockrath, N. Markovic, A. Shepard, M. Tinkham, L. Gurevich, et al., “Scanned conductance microscopy of carbon nanotubes and lambda-DNA”, Nano Lett. 2(3), 187 (2002). [8] C. Staii, A. Johnson, and N. Pinto, “Quantitative analysis of scanning conductance microscopy”, Nano Lett. 4(5), 859 (2004). [9] J. Heo and M. Bockrath, “Local electronic structure of single-walled carbon nanotubes from electrostatic force microscopy”, Nano Lett. 5(5), 853 (2005). [10] T. Jespersen and J. Nygård, “Charge trapping in carbon nanotube loops demonstrated by electrostatic force microscopy”, Nano Lett. 5(9), 1838 (2005). [11] T. Jespersen and J. Nygård, “Exposure of carbon nanotubes in dielectric layers using electrostatic force microscopy”, Submitted to Appl. Phys. Lett. (2006). [12] T. Jespersen and J. Nygård, “Probing induced defects in individual carbon nanotubes using electrostatic force microscopy”, Appl. Phys. A. (to appear) (2007). 114 [13] T. Jespersen, P. Lindelof, and J. Nygård, “Characterization of carbon nanotubes on insulating substrates using electrostatic force microscopy”, Proceedings of the International Winterschool on Electronic Properties of of Novel Materials 786, 135 (2005). [14] S. Reich, C. Thomsen, and J. Maultzsch, Carbon Nanotubes: Basic Concepts and Physical Properties (WILEY-VCH, 2004). [15] J. Nygård, Experiments on Mesoscopic Electron Transport in Carbon Nanotubes, Ph.D. thesis, University Of Copenhagen (2000). [16] K. Grove-Rasmussen, Electronic Transport in Single Wall Carbon Nanotubes, Ph.D. thesis, University Of Copenhagen (2006). [17] M. Dresselhaus and P. Eklund, “Phonons in carbon nanotubes”, Adv. Phys. 49(6), 705 (2000). [18] Group of A.Zettl, University of California, Berkeley. http://www.physics.berkeley.edu/research/zettl/projects/imaging.html. [19] L. Zheng, M. O’connell, S. Doorn, X. Liao, Y. Zhao, et al., “Ultralong single-wall carbon nanotubes”, Nat. Mater. 3(10), 673 (2004). [20] Z. Li, Z. Tang, H. Liu, N. Wang, C. Chan, et al., “Polarized absorption spectra of single-walled 4 angstrom carbon nanotubes aligned in channels of an alpo4-5 single crystal”, Phys. Rev. Lett. 8712(12), 127 401 (2001). [21] M. Bockrath, D. Cobden, P. Mceuen, N. Chopra, A. Zettl, A. Thess, and R. Smalley, “Single-electron transport in ropes of carbon nanotubes”, Science 275(5308), 1922 (1997). [22] S. Tans, M. Devoret, R. Groeneveld, and C. Dekker, “Electron-electron correlations in carbon nanotubes”, Nature 394(6695), 761 (1998). [23] S. Tans, A. Verschueren, and C. Dekker, “Room-temperature transistor based on a single carbon nanotube”, Nature 393(6680), 49 (1998). [24] J. Nygård, D. Cobden, M. Bockrath, P. Mceuen, and P. Lindelof, “Electrical transport measurements on single-walled carbon nanotubes”, Appl. Phys. A 69(3), 297 (1999). [25] J. Hafner, C. Cheung, T. Oosterkamp, and C. Lieber, “High-yield assembly of individual single-walled carbon nanotube tips for scanning probe microscopies”, J. Phys. Chem. B 105(4), 743 (2001). [26] R. Feynman, The Feynman lectures on physics (Addison Wesley Longman, 1970). 115 [27] N. Wilson and J. Macpherson, “Enhanced resolution electric force microscopy with single-wall carbon nanotube tips”, J. Appl. Phys. 96(6), 3565 (2004). [28] S. Kalinin, D. Bonnell, M. Freitag, and A. Johnson, “Carbon nanotubes as a tip calibration standard for electrostatic scanning probe microscopies”, Appl. Phys. Lett. 81(4), 754 (2002). [29] Estimates are based on typical values measured for ∼ 1µm long devices made on the CVD-grown nanotubes. The conductivities were typically around & 0.1e2 /h for metallic tubes and & 0.001e2 /h for tubes showing semiconducting behavior. [30] M. Biercuk, S. Garaj, N. Mason, J. Chow, and C. Marcus, “Gatedefined quantum dots on carbon nanotubes”, Nano Lett. 5(7), 1267 (2005). [31] J. Park, Y. Yaish, M. Brink, S. Rosenblatt, and P. Mceuen, “Electrical cutting and nicking of carbon nanotubes using an atomic force microscope”, Appl. Phys. Lett. 80(23), 4446 (2002). [32] J. Robinson, E. Snow, S. Badescu, T. Reinecke, and F. Perkins, “Role of defects in single-walled carbon nanotube chemical sensors”, Nano Lett. 6(8), 1747 (2006). [33] Q. Wang, T. Corrigan, J. Dai, R. Chang, and A. Krauss, “Field emission from nanotube bundle emitters at low fields”, Appl. Phys. Lett. 70(24), 3308 (1997). [34] C. Zhi, X. Bai, and E. Wang, “Enhanced field emission from carbon nanotubes by hydrogen plasma treatment”, Appl. Phys. Lett. 81(9), 1690 (2002). [35] Y. Liu, L. Liu, P. Liu, L. Sheng, and S. Fan, “Plasma etching carbon nanotube arrays and the field emission properties”, Diam. Rel. Mat. 13(9), 1609 (2004). [36] Commercially available www.cnanotech.com. from Carbon Nanotechnologies Inc. [37] M. Fuhrer, B. Kim, T. Durkop, and T. Brintlinger, “High-mobility nanotube transistor memory”, Nano Lett. 2(7), 755 (2002). [38] M. Radosavljevic, M. Freitag, K. Thadani, and A. Johnson, “Nonvolatile molecular memory elements based on ambipolar nanotube field effect transistors”, Nano Lett. 2(7), 761 (2002). 116 [39] W. Kim, A. Javey, O. Vermesh, O. Wang, Y. Li, and H. Dai, “Hysteresis caused by water molecules in carbon nanotube field-effect transistors”, Nano Lett. 3(2), 193 (2003). [40] J. Kong, N. Franklin, C. Zhou, M. Chapline, S. Peng, K. Cho, and H. Dai, “Nanotube molecular wires as chemical sensors”, Science 287(5453), 622 (2000). [41] P. Collins, K. Bradley, M. Ishigami, and A. Zettl, “Extreme oxygen sensitivity of electronic properties of carbon nanotubes”, Science 287(5459), 1801 (2000). [42] M. Kruger, M. Buitelaar, T. Nussbaumer, C. Schonenberger, and L. Forro, “Electrochemical carbon nanotube field-effect transistor”, Appl. Phys. Lett. 78(9), 1291 (2001). [43] W. Kim, H. Choi, M. Shim, Y. Li, D. Wang, and H. Dai, “Synthesis of ultralong and high percentage of semiconducting single-walled carbon nanotubes”, Nano Lett. 2(7), 703 (2002). [44] B. Terris, J. Stern, D. Rugar, and H. Mamin, “Contact electrification using force microscopy”, Phys. Rev. Lett. 63(24), 2669 (1989). [45] C. Schonenberger and S. Alvarado, “Observation of single chargecarriers by force microscopy”, Phys. Rev. Lett. 65(25), 3162 (1990). [46] M. Guggisberg, M. Bammerlin, C. Loppacher, O. Pfeiffer, A. Abdurixit, et al., “Separation of interactions by noncontact force microscopy”, Phys. Rev. B 61(16), 11 151 (2000). [47] J. Colchero, A. Gil, and A. Baro, “Resolution enhancement and improved data interpretation in electrostatic force microscopy”, Phys. Rev. B 64(24), 245 403 (2001). [48] J. Stern, B. Terris, H. Mamin, and D. Rugar, “Deposition and imaging of localized charge on insulator surfaces using a force microscope”, Appl. Phys. Lett. 53(26), 2717 (1988). [49] P. Ajayan, O. Stephan, C. Colliex, and D. Trauth, “Aligned carbon nanotube arrays formed by cutting a polymer resin-nanotube composite”, Science 265(5176), 1212 (1994). [50] M. Moniruzzaman and K. Winey, “Polymer nanocomposites containing carbon nanotubes”, Macromolecules 39(16), 5194 (2006). [51] V. Bliznyuk, S. Singamaneni, R. Kattumenu, and M. Atashbar, “Surface electrical conductivity in ultrathin single-wall carbon nanotube/polymer nanocomposite films”, Appl. Phys. Lett. 88(16), 164 101 (2006). 117 [52] J. Foster, S. Singamaneni, R. Kattumenu, and V. Bliznyuk, “Dispersion and phase separation of carbon nanotubes in ultrathin polymer films”, J. Colloid Interface Sci. 287(1), 167 (2005). [53] C. Stephan, T. Nguyen, M. de la chapelle, S. Lefrant, C. Journet, and P. Bernier, “Characterization of singlewalled carbon nanotubes-PMMA composites”, Synth. Met. 108(2), 139 (2000). [54] Z. Jia, Z. Wang, C. Xu, J. Liang, B. Wei, D. Wu, and S. Zhu, “Study on poly(methyl methacrylate)/carbon nanotube composites”, Mater. Sci. Eng. A 271(1-2), 395 (1999). [55] G. Sacha, C. Gmez-Navarro, J. Senz, and J. Gmez-Herrero, “Quantitative theory for the imaging of conducting objects in electrostatic force microscopy”, Appl. Phys. Lett. 89, 173 122 (2006). [56] E. Minot, F. Kelkensberg, M. van Kouwen, J. van Dam, L. Kouwenhoven, et al., “Single quantum dot nanowire LEDs”, Nano Lett. p. ASAP (2007). 118 Appendix A InAs nanowire growth NBI73 The InAs wires which have been studied in this thesis was grown by MBE on a (100) GaAs substrate. A thin Au film was deposited with a substrate temperature of 760 ◦ C(Au source at 1300 ◦ C) , followed by 15 min of GaAs growth (substrate temp. 580 ◦ C, As source 200 ◦ C, Ga source 910 ◦ C) and 30 min of InAs growth (substrate temp. 470 ◦ C, As source 200 ◦ C, In source 887 ◦ C). The growth was developed and performed by Martin Aagesen (manuscript in preparation). Figure A.1 shows additional images of the growth product (see section 2.2). 119 Figure A.1: SEM images of the nanowire growth NBI73 used for the devices in chapter 2. For some of the wires gold nano particles can be observed in the ends. 120 Appendix B Details of nanowire device fabrication. This appendix provides the details for the fabrication of nanowire devices as well as a discussion of the precautions needed to allow a reasonable fraction of (extremely sensitive) devices to reach the stage of low-temperature measurements. The fabrication follows along these lines: • The starting point is a substrate wafer of highly doped silicon (which will stay conducting to dilution fridge temperatures) capped with 500 nm of insulating SiO2 . The final device will be supported on this substrate and the conducting silicon will act as a global electrostatic (back) gate. The substrate is cut into manageable sizes of ∼7×10mm and prepared with a grid of metal alignment marks fabricated by standard e-beam lithography (for parameters, see below) and subsequent metal evaporation (Cr/Au, 10/40 nm). The grid is needed for finding and contacting nanowires as described below, and a grid-size of ∼10 − 20 µm was found suitable. Furthermore, large metal bonding pads were fabricated around each area of alignment marks, using standard UV-lithography and (Cr/Au, 20/180 nm) metal evaporation. • To transfer the nanowires from the high density growth-wafer to the device substrate a very simple mechanical approach was employed. A small (4×4 mm) piece of the growth substrate was placed face-down on the device substrate and moved gently back and forth a few times while applying as little downwards pressure as possible. In this way some wires break off the growth substrate and stick to the sample. The result can easily be checked in a high resolution optical microscope and the process can be repeated until the desired density of wires is achieved. Figure 2.7(a) shows a typical optical microscopy image of InAs nanowires deposited in an alignment grid. • Using either the optical images or SEM images a contact pattern is 121 now designed using the alignmentmarks as reference. SEM images have the advantage of higher resolution and are therefore required for advanced device geometries, however, it is faster to obtain the optical images and furthermore, they can be obtained after spinning on the e-beam resist. This is important since the wires, in some cases, tend to move when spin coating the sample with resist, leading to subsequent misalignment. This is avoided by obtaining the optical images for alignment after spinning on the resist and since we were only interested in making simple two-terminal devices the limited resolution was not a concern and the optical method was adopted in most cases. • The contacts are defined by standard e-beam lithography: The chip is coated with resist layers of 6% copolymer and 4% PMMA by spinning at 4000 rpm for 45 s and baked on a hotplate at 185 ◦ C for 90 s. The lithography is performed using a JEOL JSM-6320F scanning electron microscopy running Elphy lithography software using an exposure voltage of 30 keV and a current of ∼20 pA. The patterns were developed for 60 s in MIBK:IPA (1:3) and flushed in IPA. • Prior to metal evaporation the samples were ashed for 5 − 10 s in an oxygen plasma etch to remove resist residues from the wire surface. This was followed by a brief ∼5 s wet etch in buffered hydrofluoric acid to remove the oxide from the nanowire surface (cf. TEM image, Fig. 2.6). The sample was then flushed with deionized water and blow-dried with N2 and immediately loaded in the load-lock of the metal evaporation chamber. In order to prevent (as far as possible) re-oxidation of the wire, efforts were made to bring down the time between the end of the HF etch and loading in the evaporation chamber. Usually the load-lock was evacuated within 2 minutes of the HF etch. Based on the relatively few devices (< 20) which were electrically characterized it seems that omitting the plasma and HF etch result in devices with large contact barriers (devices in the Coulomb blockade regime). In contrast, by including the cleaning procedure, better contact is achieved and the device transparency can be tuned by the gate (section 2.1). For the contacts either a Ti/Au (∼10/50 nm) bilayer was used (for normal-metal contacts) or a Ti/Al/Ti (∼10/60/10 nm) trilayer for superconducting contacts (see section 2.7). Lift-off was performed in acetone and Fig. 2.7(b) shows an SEM micrograph of a typical finished device. • To achieve good contact to the substrate (back gate) the back of the sample was metallized (Cr/Au, 10/40 nm) and the sample glued to a chip-carrier using a conducting silver-paste. The remaining and final step is to wire-bond the sample to the chip-carrier. An ultrasonic ball bonder (Kliche & Soffa 4124) was used for this purpose, how122 ever, special care was needed since the nanowire devices were found to be extremely sensitive to electrical/static shocks (much more so than nanotube based devices). Hundreds of devices were destroyed in the bonding process and the subsequent handling, but taking the following precautions enabled a reasonable number of devices to reach the measurement step: *) Always ensuring extensive grounding of bonding machine (and user). *) Using an ionizing fan to limit static buildup. *) Before bonding from the chip-carrier to the sample, all pads on the chip-carrier were connected by bonding wire-loops large enough to be subsequently removed without the use of microscope (after mounting the sample in the cryostat). *) Using the bonder in ”manual” mode, i.e., preventing automatic electrical cutting of the bonding wire. *) After bonding the two leads of a device, the SEM was used to assess if the wire was still intact. In this case the sample was mounted in the cryostat - bonding additional devices was sometimes/often found to destroy all previously bonded devices. *) Finally, the sample was mounted in the grounded cryostat and the bonding-loops removed with tweezers (user grounded). Note the absence of a room-temperature probingstep; it has been found to be nearly impossible to use a probe-station to characterize the devices before bonding: in such attempts nearly all devices were lost. 123 Appendix C Type-1 devices in the closed-dot regime 5.0 Vsd [mV] 2.5 0.0 -2.5 -5.0 -7.5 -7.0 Vg [V] -6.5 Figure C.1: Stability diagram measured in the closed-dot regime of a type-1 device measured at T = 300 mK (darker = more conductive). The arrows indicate the level spacings which are generally larger than for the type-2 devices treated in section 2.4. The measurement has been corrected for a number of gate-switches. See discussion in the main text. 124 Appendix D Additional Kondo data Figure D.1 provides additional data measured on the same device as discussed in connection to Fig. 2.31 (section 2.7). The data illustrates the possibility of a large TK Kondo peak (TK & ∆) surviving the transition to superconducting leads (see figure caption). Figure D.2 shows a stability diagram measured at T = 50 mK for a device with a single-wall carbon nanotube contacted by superconducting leads in much the same way as the nanowire devices presented in this thesis. The device fabrication and the measurement were carried out at the Niels Bohr Institute by K. Grove-Rasmussen and H.I. Jørgensen - for details see K. Grove-Rasmussen et. al. ”Interplay between supercurrent and Kondo effect in single wall carbon nanotube Josephson junctions” cond-mat/0601371. The pronounced ∆-peaks are observed along a suppressed Kondo ridge which shows that the observation discussed in section 2.7 is not related to the nanowire system, but is rather a general phenomenon of Kondo-dots connected to superconducting leads. 125 3 3 800 mK Vsd [mV] Vsd [mV] κ 0 b 300 mK 0 (a) (c) -3 -3 Vg Vg 9.0 5.0 G [e2/h] G [e2/h] κ b 0.0 (b) -0.6 7.0 5.0 (d) 0.0 Vsd [mV] 0.6 0.5 1.0 T [K] 2.0 2∆ ∆ 0.2 0.2 Vsd [mV] Vsd [mV] Figure D.1: (a) Stability diagram for T = 800 mK, i.e., above the superconducting transition temperature. A clear Kondo ridge is seen as also emphasized in (b) showing line traces through the center of the Kondo ridge and the adjacent CB valley (cf. 2.31(b)). (c) Corresponding stability diagram for T = 300 mK. The 2∆ peaks are clearly seen and the Kondo ridge remains. (d) Shows the temperature dependence of the valley conductance exhibiting an increase through Tc rather than a decrease as for Kondo valleys with lower TK . The reason for the very high conductance (even in the normal state) is not understood. 0.0 0.0 -∆ -2∆ -0.2 -0.2 dI/dV [e2/h] -5.0 -4.8 Vg [V] -4.9 -5.0 -4.8 Vg [V] -4.9 1.4 0.0 Figure D.2: Measurement from a CNT-based quantum dots contacted by superconducting leads (see text). For normal leads (left) a Kondo resonance is observed which is suppressed for superconducting leads (right). Along the suppressed ridge the enhanced ∆-peaks are clearly observed. Colorscale identical for the two plots. 126 Appendix E EFM parameters The various parameters used when scanning in EFM mode can readily be found in the screen-shot from the SPM controller software shown in Fig. E.1. 127 Figure E.1: Screen-shot of a typical EFM run showing the typical parameters and settings used. For greater magnification, view the electronic version of the thesis. 128 Appendix F Publication list During the course of the project the following papers have been published (or are in preparation): • Williams J.R., Sand-Jespersen T., Marcus C.M., et. al., ”Weak localization and conductance fluctuations in mesoscopic graphene”, In preparation (2007) • Sand-Jespersen T., Grove-Rasmussen, K., Ingerslev, H., Aagesen, M., Sørensen, C., Lindelof, P.E., Paaske, J., Andersen, B.M., Flensberg, K., Nygård J. ”Effect of Kondo Correlations on the Subharmonic Gap Structure of a Quantum Dot Coupled to Superconductors”, In preparation (2007) - section 2.7. • Sand-Jespersen T., Nygård J. ”3D-Mapping of Individual Carbon Nanotubes in Polymer/Nanotube Composites using Electrostatic Force Microscopy”, Submittet to Appl. Phys. Lett. (2007) - section 3.6. • Sand-Jespersen T., Nygård J. ”Exposure of Carbon Nanotubes in Dielectric Layers using Electrostatic Force Microscopy”, Submittet to Appl. Phys. Lett. (2006) - section 3.1 partly. • Sand-Jespersen T., Nygård J. ”Probing Induced Defects in Individual Single Wall Carbon Nanotubes using Electrostatic Force Microscopy”, Accepted, Appl. Phys. A. (2007) - section 3.4. • Sand-Jespersen T., Aagesen, M., Sørensen, C., Lindelof, P.E., Nygård J. ”Kondo effect in tunable semiconductor nanowire quantum dots”, Phys. Rev. B. 74, 233304 (2006) - section 2.5. • Sand-Jespersen T., Nygård J. ”Charge Trapping in Carbon Nanotube Loops Demonstrated by Electrostatic Force Microscopy”, Nano Lett. 9, 1838 (2005) Cover Story - section 3.5. • Sand-Jespersen T., Lindelof P.E., Nygård J. ”Characterization of Carbon Nanotubes on Insulating Substrates using Electrostatic Force Microscopy”, AIP Conference Proceedings,786, 135 (2005) - section 3.1 partly. 129