Exploring matter at the atomic level

1 Why transmission electron microscopy for nanoelectronics?

Exploring matter with increasing spatial resolution is not only the answer to our pure scientific curiosity, it is also a crucial challenge aimed to improve our capability to control physical properties of materials used for a variety of technological applications. In the field of microelectronics, for instance, one of the main motivations that has been driving our effort to improve performances of microscopy techniques is actually related to the consequences of a pure economic law, the one introduced by Gordon Moore, the co-founder of Intel, back to 1965. Moore’s law (fig. 1a) states that the price per million of transistors or, similarly, the price per million of instructions processed per second, reduces by 35% per year, following a logarithmic trend. This law has indeed accompanied the growth of the so-called information communication society in the form we experience it today, dominated by a multitude of microelectronic devices that help us to transmit and elaborate huge amount of data. It is not surprising that the worldwide revenue from semiconductor industry has been constantly rising in the last five decades, by approaching last year the level of 400 billions of US dollars.

The effectiveness of Moore’s law strictly depends on our capability to progressively shrink the size of a single transistor fabricated in a silicon chip (fig. 1b). In terms of transistor channel length (i.e. the distance between the source and the drain of the device) we are now facing the 10 nm node. In order to minimize capacitance cross-talk effects among tremendously packaged elementary devices realized on the same chip, the microelectronics industry has abandoned the classical planar architecture of the metal oxide semiconductor (MOS) technology, and the field effect transistor (FET) is today a three-dimensional structure (fig. 2) based on a silicon fin containing source and drain in a few tens of nanometers.

It is relevant to emphasize that this technological progress has been made possible, to a significant extent, thanks to our capability to explore matter with an increasing level of spatial resolution, by introducing emerging and more powerful microscopy techniques, moving from optical, to transmission electron microscopy (TEM) and, more recently, to spherical aberrationcorrected scanning-transmission electron (STEM) microscopy (fig. 3).

In order to highlight the resolution that a last-generation STEM microscope can achieve, fig. 4a shows the typical picture of a single-crystal silicon sample, as observed, along its (110) axis, by a pure conventional TEM. The image is the result of the interference of a coherent parallel electron beam illuminating the sample (fig. 4b). In this micrograph the contrast is related to the periodical distribution of the electrostatic potential in the lattice. Conversely, in the last generation of STEM microscopes, such as the state-of-the-art instrument of the Beyond-Nano facility of the National Research Council of Italy, a sub-ångström electron probe is focused and scanned over the sample (fig. 5b). Such an extremely narrow focusing is possible thanks to the use of complex electro-optical lenses that, in modern electron microscopes, significantly reduce the disc confusion produced by spherical aberration, a well-known effect since the invention of the first optical telescope by Galileo Galilei.

The contrast in the magnified micrograph of the sample, shown in fig. 5a, is now modulated by the intensity of electrons scattered by individual columns of atoms that is proportional to the square of the atomic number. Under these conditions, interference effects are minimized and the position of any individual atom in the lattice can be detected very precisely. The final micrograph we get corresponds to the real structure of the direct silicon crystalline lattice and any interference point of the conventional TEM picture (fig. 4a) is now replaced by the characteristic Si dumbbell, expected to be observed along the (110) orientation (fig. 5c).

2 Novel materials: what beyond silicon technology?

Such high-spatial-resolution techniques are extremely important to investigate matter at the atomic level and, from the technological point of view, in the field of microelectronics, to effectively face the ultimate limit of the silicon technology. Indeed, as the channel length of the fin FET device architecture will reduce to a distance below 10 nm, direct quantum-mechanical tunneling of carriers, from the source to the drain, may prevent any deterministic control of the electrical characteristics of the device. This is believed to be the major bottleneck for further scaling of this technology, and in many cases the solution to overcome the intrinsic obstacles of the silicon technology is to introduce novel materials, device architectures, and complex heterostructures.

2.1 Phase change materials

In the field of non-volatile memory devices, for instance, a promising approach is the one based on the use of phase change materials (PCM), such as the alloy consisting of a mixture of germanium, antimony and tellurium (GST). These materials are characterized by strong differences in their electrical conductivity between the crystal and the amorphous phase. The transition from the amorphous to the crystalline state, and vice versa, can be achieved by sharp temperature changes induced by a proper electrical current pulse, thus allowing the fabrication of non-volatile memory devices whose logic state, “0” or “1”, is controlled by a conductivity measurement. As the electrical pulse goes up to the “set” value (fig. 6a), the temperature of the GST film, initially in its amorphous phase, rapidly increases (fig. 6b) to the range of values triggering fast crystallization. A subsequent lower pulse, used to probe the logic state of the device, finds the film conductivity in its highest value (fig. 6c), characteristic of the crystal phase. A new voltage pulse can now be applied to increase the temperature above a value that induces solid-liquid transition (“reset” level in fig. 6a). By switching off the reset pulse, the GST film will solidify by fast quenching in the amorphous phase and the subsequent reading will find the material in its lowest conductivity state, corresponding to the “0” logic state.

Actually, the crystalline structure of the GST film can be quite complex. As shown in the micrographs of fig. 7, the material can be found in its rock-salt cubic phase (fig. 7a) or in the trigonal one (fig. 7c), depending on the deposition temperature and on the substrate we use to grow the film on. The aberration-corrected STEM technique gives the opportunity to get atomic contrast due to the dependence – as already mentioned – of the electron scattering cross section on the square of the atomic number (Z-contrast technique). Under these conditions, the atoms with larger atomic number appear brighter in the micrograph and the position of germanium, antimony and tellurium atoms in the lattice can be perfectly distinguished and directly compared with simulation (figs. 7b and 7d).

The micrograph of fig. 7c emphasizes the presence of structural gaps in the trigonal lattice, periodically distributed along the layer depth, with a periodicity that can be controlled by modifying the deposition parameters. It has been found that these gaps are generated between two consecutive planes of tellurium atoms, the green spheres in the micrograph of fig. 7c, whose tetrahedral bonds are all oriented to the germanium and antimony atoms lying above or beneath in the stack. Consequently, these adjacent tellurium planes, facing each other, are kept bound only by weak van der Waals forces.

The electron microscopy study is extremely relevant from the electronic transport point of view. Indeed, it has been recently shown that multi-layered crystalline PCM, known as interfacial PCM (iPCM), can have improved functional properties, with a 95% reduction in power consumption in comparison to classical PCM. This is probably due to the fact that, while the high resistance (amorphous) phase is usually obtained by quenching the molten phase, in multilayered GeTe/Sb₂Te₃ structures (iPCM), instead, a solid-solid phase transition occurs. Moreover, it has been reported that iPCM has a 2000% electrical-induced magnetoresistance at room temperature. This could allow designing of conceptually novel non-volatile memory devices with the combined merits of phase change and magnetic memories, as well as new types of magnetic sensors. Therefore, a comprehensive understanding of the atomic rearrangement, thanks to the atomic-resolution STEM technique, is relevant for iPCM development and their data retention, since the stacking sequences and their spacing can play a key role in the switching mechanism.

2.2 Two-dimensional materials

Since 2010, after the Nobel Prize assigned to Novoselov and Geim, graphene and two-dimensional (2D) materials have attracted an increasing interest, for both the scientific and technological point view. Atomic-resolution electron microscopy, combined with integrated spectroscopic techniques (electron energy loss and/or energy dispersion X-ray), is a crucial methodology for investigating structural and chemical properties of these materials. As an example, fig. 8 shows atomically resolved plan view micrographs of free-standing graphene (fig. 8a), exfoliated phosphorene layer (fig. 8b), and of pristine molybdenum disulfide (MoS₂, fig. 8c). In the latter image STEM microscopy highlights the effectiveness of the Z-contrast methodology in discriminating the location of any individual sulfur and molybdenum atom of the MoS₂ honeycomb structure.

Thanks to their intrinsic thin body, these materials are considered quite promising in the field of microelectronics for the possibility to use them to maximize the gate modulation efficiency in utrashort-channel transistors. Indeed, the thickness of these materials extends just one atom. However, in spite of the many superlative characteristics, the main drawback of using graphene in the transistor technology is the lack of an energy gap between the conductive and the valence band. This problem can be overcome by transition metal dichalcogenides, such as molybdenum disulfide, which instead exhibits a semiconducting behavior with an energy gap of 1.8 eV in its layered structure. Even more important, the high effective mass of carriers in molybdenum disulfide can minimize direct source-drain tunneling, making this material very interesting for next-generation transistor technology.

From the electrical transport point of view pristine molybdenum disulfide exhibits the typical behaviour of an n-type semiconductor. This can be seen, for instance, by measuring the local current, at the nanoscale level, by using conductive atomic force microscopy where a sharp metal tip (having a contact radius of about 10 nm) is scanned over MoS₂ multilayers flakes, exfoliated on SiO₂, as depicted in the schematic inset shown in fig. 9a. A bias voltage is applied to the metal contact surrounding the film and the current-voltage characteristics are collected in any precise position of the tip. These electrical characteristics (continuous curves in the linear plot of fig. 9a) confirm that electrical transport can be described in terms of a Schottky barrier contact on an n-type semiconductor, with a current onset at low forward bias and a high leakage current under reverse bias.

It was recently discovered that, when exposing the MoS₂ film to an oxygen plasma treatment, its conduction behaviour progressively converts from n- to p-type. This is demonstrated by the change of the I-V characteristics plotted in fig. 9b. In addition to n-type Schottky contact curves (that now show a larger spread compared to the case of pristine MoS₂), some curves (drawn in red in fig. 9b) exhibit the typical behaviour for Schottky contacts on a p-type semiconductor, i.e. the onset of a negative current for negative voltages and negligible current for positive voltages.

This result is technologically relevant for the opportunity to access to both types of electrical conduction, n or p, by properly processing the MoS₂ material on a same device, and sheds light on the possibility of fine-tuning MoS₂ electronic properties which can find many applications for next-generation electronic/optoelectronic devices based on this material.

Once more, atomic resolution electron microscopy is essential to optimize the processing technique. Indeed, we observe that oxygen plasma does not alter the structural integrity of the 2D material, as demonstrated by the micrograph shown in fig. 10a. A modern STEM is generally equipped with electron energy loss spectrometer (EELS), a technique that, in the discussed system, provides direct evidence of the molybdenum oxidation induced by oxygen plasma. Indeed, this can be seen by observing the shift of the M3 and M2 ionization edges towards higher energy values, accompanied by the appearance of the oxygen-related K ionization edge in the EELS data (fig. 10b).

The study of the electronic properties of 2D materials is an important step in the assessment of their application capabilities. This can be carried out through the analysis of fine structures in electron energy loss spectrum, originating in the transition from a core level to unoccupied states under the dipole selection rule (in dipole scattering conditions). The excitation of core-level electrons into unoccupied orbitals provides information on chemical shifts of core-level states, as well as the fine structure in the unoccupied valence-band states. Indeed, Energy-Loss Near-Edge Structure (ELNES), coupled with scanning transmission electron microscopy (STEM) represents a powerful local probe to observe the distribution of valence states at the different core holes, which can be selected by the energy dependence of the absorption cross-section. Furthermore, direct imaging of the lattice yields information on atomic positions that can be used in theoretical models to reproduce the EELS spectra in order to obtain the density of states.

This analytical methodology was recently used to investigate, for instance, the excited states of the energy band structure of black phosphorous (BP), a semiconductor having a direct bandgap that can be varied from 0.3 eV in the bulk to 1–2 eV in the monolayer limit, due to quantum confinement. This circumstance makes this material particularly ideal for electronics and optoelectronics applications. The atomic structure of black phosphorous in cross-sectional configuration is shown by the STEM micrograph of fig. 11a. Black phosphorous atoms are arranged in puckered honeycomb layers bound together by van der Waals forces, and, in that sense, the material can be thought of as composed of a pile of phosphorene.

The experimental K-edge of black phosphorus (green curve in fig. 11b) shows different features at energies higher than the edge. Features in ELNES are related to the population of high-lying conduction states by valence electrons that have been excited by primary electrons. Electrons emerge into the vacuum from these conduction band states, producing the fine structure that, thus, is related to the empty bands of black phosphorus. The nature of these unoccupied electronic states can be unveiled by an analysis of the density of states (DOS). In fact, a comparison of the symmetry-projected DOS and EELS spectra can relate each spectral feature to transitions to specific electronic states. The phosphorous K-edge corresponds to $1s \rightarrow 3p$ single-particle transitions, according to dipole selection rules, and the 3p orbital contributions to the conduction band can be directly compared with the experimental K-edge measured by EELS.

In fig. 11b, the experimental spectrum (green curve) is directly related to the theoretical one (red curve) simulated by using the Feff 8.4 code, an ab initio self-consistent real-space multiple-scattering code used for simultaneous calculations of X-ray absorption spectra and electronic structure. The same code provides the corresponding information on partial DOS. Features in the experimental ELNES spectrum probed by EELS correspond to high values of the unoccupied partial DOS. In particular, the calculated partial DOS well reproduces experimental features at about 20 and 26 eV over the edge.

In future studies, it would be interesting to focus this investigation methodology on excited states in monolayer black phosphorus (phosphorene), in which two-dimensional discrete states at energies above the vacuum level are expected to be embedded in the three-dimensional continuum, thus turning discrete states into resonances.

2.3 The graphene/silicon carbide heterostructure

Coming back to graphene, although it still does not appear to be the most suitable material in the transistor technology (for the lack of an energy bandgap), it is becoming quite attractive in the silicon carbide technology. Silicon carbide is a wide band-gap semiconductor whose importance is increasing for applications in power electronics, a field extremely important for several applications related to the efficiency of the electrical-energy distribution and power conversion. Silicon carbide, however, is not so effective for high-frequency electronics due to the intrinsic difficulty to realize two-dimensional electron gases on it. This problem could be overcome by integrating graphene on silicon carbide. The hetero-structure consisting of epitaxial graphene grown on silicon carbide is believed to be a promising system to integrate high-power and high-frequency functions on the quite mature silicon carbide technology.

Graphene presents specific structural and electronic properties depending on the growth substrate and mechanism, which consequently have an impact on its macroscopic electrical behavior. On the Si-terminated face of a SiC substrate, in its hexagonal 4H phase (4H-SiC), the epitaxial growth of graphene needs the presence of nucleation kinks at the wafer surface. These kinks are usually favored by properly mis-cutting the SiC wafer (fig. 12a) and, in the example discussed here, ($\mathrm{0001}$) SiC wafers, cut with off-axis angle of about 8°, were used. Figure 12b shows the atomic force microscope image of such a surface, presenting tilted walls parallel to the ($\mathrm{1\bar{1}00}$) orientation, with a height of about 50 nm.

The epitaxial growth of graphene on SiC is induced by Si sublimation at high temperature (above 1200 °C). Atomic resolution STEM analysis provides the direct picture of the graphene-SiC system, as shown in fig. 13a where the cross-sectional view around one of the tilted walls on the SiC wafer surface is displayed. On the $[\mathrm{0001}]$ silicon carbide flat terrace, the micrograph shows the presence of a stacked structure (see the detail in fig. 13b), with five graphene layers spaced by a distance of 3.37 Å. The presence of a carbon buffer interfacial layer (the first one of the stack), with a smaller distance from the SiC surface (2.62 Å) is also detected.

This measurement has been found to be consistent with ab initio calculations of the equilibrium atomic distance with the silicon-atoms of the silicon-terminated ($\mathrm{0001}$) surface. By moving toward the inclined surface (fig. 13c), however, the spacing between this first carbon layer and the substrate increases to 3.46 Å, demonstrating that the buffer layer gets detached from the SiC surface.

The structural results were related to local measurements of the electrical current by using conductive atomic force microscopy (fig. 14). When synthesized on a silicon carbide ($\mathrm{0001}$) surface, epitaxial graphene undergoes a high electron doping, originating from unsaturated Si dangling bonds at the interface with the carbon buffer layer. This n-type doping is responsible of the relatively high conductivity of the epitaxial graphene layers. Along the inclined surfaces, where the carbon buffer layer gets detached, the conductivity of the graphene stack falls down, thus creating isolation walls parallel to the ($\mathrm{1\bar{1}00}$) direction. This result is particularly useful for applications, since it is possible to realize regions with high values of conductivity which are isolated from each other on the same silicon carbide wafer avoiding particularly complex photo-lithographic processes.

To strengthen these structural and transport evidences, the electronic structure was probed through atomic-scale EELS imaging and spectroscopy in the energy loss region around the K ionization edge of carbon, by using a sub-ångström electron spot focused on individual atoms of the graphene layers (fig. 15). When the electron probe is moved on the second graphene layer of the stack, the presence of the $\pi^{*}$ peak, at about 285 eV, associated with the $sp^{2}$ two-dimensional hybridization, is clearly identified in the electron energy loss spectrum (violet curve) together with a weaker signal from the $\sigma^{*}$ peak, at about 293 eV, due to the carbon in the $sp^{3}$ hybridization. By moving the spot on the interfacial carbon layer attached to the $[\mathrm{0001}]$ flat surface of silicon carbide, the height of the $\pi^{*}$ peak strongly diminishes, correspondingly the $\sigma^{*}$ signal increases (blue curve), demonstrating that this layer is far from the graphene configuration. Still following the first layer but now along the titled wall of the surface, we notice that the $\pi^{*}$ peak (the $sp^{2}$ hybridization) increases again (red curve), confirming that the carbon buffer interfacial layer delaminates, becoming quasi-free-standing graphene. This finding is of crucial importance for the local modification of the electrical characteristics of epitaxial graphene on SiC steps, since it should trigger intrinsic scattering mechanisms that are related with areas of unequal doping and unequal number of graphene layers.

It is impressive to notice, as a conclusion, that the importance of improving the resolution of electron microscopy was largely emphasized by Richard Feynman in its famous speech “there is plenty of room at the bottom”, delivered in 1959 at the annual meeting of the American Physical Society. Feynman predicted that increasing the performance of electron microscopes would produce a strong impact in science and technology of matter. He had the dream of “improving the electron microscope by a hundred times”, which meant a resolution in the range of 0.1 Å. The examples discussed here demonstrate that Feynman’s dream has been successfully fulfilled.