Semir Tulića,
Thomas Waitza,
Oleksandr Romanyukb,
Marián Vargab,
Mária Čaplovičovác,
Gerlinde Hablerd,
Viliam Vretenárc,
Mário Kotlárc,
Alexander Kromkab,
Bohuslav Rezekbe and
Viera Skákalová*ac
aPhysics of Nanostructured Materials, Faculty of Physics, University of Vienna, Boltzmanngasse 5, 1090 Vienna, Austria. E-mail: viera.skakalova@univie.ac.at
bInstitute of Physics, Czech Academy of Sciences, Cukrovarnická 10, Prague 6, Czech Republic
cSlovak University of Technology, Centre for Nanodiagnostics, Vazovova 5, 812 43 Bratislava, Slovakia
dDepartment of Lithospheric Research, University of Vienna, Althanstrasse 14, 1090 Vienna, Austria
eFaculty of Electrical Engineering, Czech Technical University, Technická 2, Prague 6, Czech Republic
First published on 26th February 2020
Nanocrystalline diamond (NCD) films grown on Si substrates by microwave plasma enhanced chemical vapor deposition (MWPECVD) were subjected to Ni-mediated graphitization to cover them with a conductive layer. Results of transmission electron microscopy including electron energy-loss spectroscopy of cross-sectional samples demonstrate that the oxide layer on Si substrates (∼5 nm native SiO2) has been damaged by microwave plasma during the early stage of NCD growth. During the heat treatment for graphitizing the NCD layer, the permeability or absence of the oxide barrier allow Ni nanoparticles to diffuse into the Si substrate and cause additional solid-state reactions producing pyramidal crystals of NiSi2 and SiC nanocrystals. The latter are found impinged into the NiSi2 pyramids but only when the interfacial oxide layer is absent, replaced by amorphous SiC. The complex phase morphology of the samples is also reflected in the temperature dependence of electrical conductivity, where multiple pathways of the electronic transport dominate in different temperature regions. We present models explaining the observed cascade of solid-state reactions and resulting electronic transport properties of such heterostructures.
Fig. 1d shows Raman spectra of the pristine NCD-1p (black) and the annealed NCD-1 (red) using the excitation wavelength of 488 nm. Besides less defined wide features, the spectrum of pristine NCD-1p contains a distinct diamond Raman active mode at 1330 cm−1 related to the sp3-bonded carbon phase,11 the Raman spectrum of the processed NCD-1 (red line) is dominated by the signatures of graphite: the G and 2D peaks located at 1579 cm−1 and 2718 cm−1, respectively, and the dispersive D peak at 1355 cm−1 originating from a disordered sp2-phase.12 The small value of the ratio 2D/G < 0.5 indicates the presence of a relatively thick layer of graphite; however, a small diamond peak is still resolvable. The presence of graphite can explain the continuous surface layer with low SE signal intensity shown in Fig. 1b. Using the excitation wavelength 633 nm we also identified additional weak and broad Raman peaks located between 200 cm−1 and 400 cm−1 and assigned to NiSi2 (inset of Fig. 1d).13 These signals were not detected on pristine NCD and are thus likely related to the bright rectangles located below the sample surface and visible only in the BSE image (Fig. 1c).
In order to analyse the subsurface structure, TEM investigations of cross-sectional samples have been performed. The TEM image in Fig. 2a shows the arrangement of graphite layers: well-ordered (0001) basal-planes of graphite planes emanate from the diamond grain on the left with an orientation almost perpendicular to the diamond/graphite interface, but parallel to the surface of the substrate. Fig. 2b represents a sketch of the diamond/graphite–SiO2/Si configuration imaged in Fig. 2a.
Such structures appear rather unique. They are attributed to the Ni-mediated catalytic process reported recently.10 Upon heating to T = 1073 K the thin Ni film undergoes solid-state dewetting and decomposes into nanoparticles. Mobile Ni nanoparticles penetrate along the diamond grain boundaries into the diamond layer and, on their way, catalytically release C atoms from diamond forming graphite. Covalent bonding between diamond and oriented planes of graphite was established during the catalytic transformation due to open dangling bonds, left behind the proceeding Ni particle, which are saturated with the carbon atoms released from the Ni particle. Compared to prior work, there are three distinguishable features in the present case. Although the thickness of deposited Ni was the same (20 nm) in both experiments: (i) the high ratio of graphite relative to the retained diamond; (ii) the well-ordered crystal structure of graphite; (iii) Ni particles are rarely observed in the NCD/graphite layer. These distinguished characteristics are attributed to the lower thicknesses of the NCD (∼70 nm) and SiO2 (∼5 nm) layers here, compared to the prior work.10 The key influence of these two factors on Ni activity is revealed by the analysis in Fig. 3.
The TEM image in Fig. 3a shows an overview of NCD-1 in cross-section. Protective layers of Au and Pt, as required for focused ion beam (FIB) preparation, appear darker in the bright-field (BF) image, whereas the layer above comprising retained grains of the NCD film and graphite appear brighter; within the carbon layer we also find a few individual Ni particles appearing darker than carbon in the BF image. Between the carbon layer and the Si substrate there is a more or less continuous 5 nm thin SiO2 layer with similar grey value as carbon in the BF image. Above the straight Si/SiO2 interface, dark pyramids impinge into the single crystal Si. Still, Fig. 3b shows a disruption of the SiO2 layer at the centre of the pyramid (marked by the arrow). We suppose that along such disruptions in the thin SiO2 layer, Ni can migrate into the Si substrate. A subsequent reaction with Si then forms the NiSi2 single crystal pyramids, chemical composition of which has been identified by Raman and EDX spectroscopy. The NiSi2 pyramids show almost a perfect epitaxy and a well-defined orientation relationship with the Si substrate. Their bases, appearing as rectangles in the BSE SEM image in Fig. 1c, are oriented parallel to the (001) plane, and the sides parallel to {111} planes of the silicon crystal. The STEM image in the inset of Fig. 3b demonstrates this relationship between the NiSi2 nanocrystals and the Si single crystal lattice. The atomic structure of NiSi2/Si system has been a subject to TEM and STEM investigations for some time.14–18 The surface reactivity of Ni with Si forming epitaxial NiSi and NiSi2 within a Si crystal was already explained by a mechanism of a solid-state reaction.19–23 More recent publications24–26 describe a similar process also in the presence of native SiO2 separating Ni films from (001) silicon substrates. They concluded that, since the energy barrier for metal diffusion through non-stoichiometric native SiO2 is relatively low,24,25 Ni can penetrate through and react with Si forming a NiSi2 crystal. The mechanism of decomposition of native SiO2 after thermal treatment was closely studied in several works.25–27 When temperatures reach 1173 K, Si at the SiO2/Si interface starts reacting with the thin SiO2 film to form SiO vapor. The SiO molecules leave the surface reducing the density of SiO2. Consequently, the generation of voids allows the exposure of the Si surface to Ni. Such a supposed void channel through the SiO2 layer is marked below the centre of one of the NiSi2 pyramids in Fig. 3b.
The HAADF signal of samples increases with the atomic number of their chemical elements and the sample thickness. The small crystals at the pyramid bases appear brighter than carbon below but darker than NiSi2 (Fig. 4b and d). EELS confirms the composition of the crystals as SiC; whereas HRTEM images reveal the lattice structure of cubic 3C-SiC; frequently stacking faults and nanotwins are visible in 3C-SiC adjacent to diamond (indicated with A in Fig. 4c). HRTEM also shows a well-defined crystallographic orientation relationship of the SiC nanocrystals with the NiSi2 lattice structure (their (002) and {111} planes are almost parallel to each other, Fig. 4c). However, while SiC crystals show well-developed facets in contact with NiSi2, their interface geometry with C phases appears rather diffused.
Fig. 5 HRTEM images of the phases occurring near the base of NiSi2 pyramids formed in samples (a) NCD-1 and (b) NCD-2. |
Fig. 6 (a and b) HAADF STEM images of two interfacial areas between the Si substrate and the carbon layer in NCD-1; (c and d) superposition of EELS maps of K-edges of Si (blue), C (red), O (green), and Ni L-edge (yellow). The white dashed line in (c) marks the position of the profile shown in Fig. 8a. |
The composite images in Fig. 6c and d show a superposition of EELS maps of K-edges of Si (blue), C (red), and O (green) as well as the Ni L-edge (yellow). They correspond to two different interfacial areas shown as HAADF STEM images in Fig. 6a and b, respectively. The left panels (Fig. 6a and c) display the diamond/Si interface, whereas the right panels (Fig. 6b and d) the diamond/NiSi2 interface. Besides some Ni nanoparticles that have penetrated into the interfacial layer, both areas show a rather continuous green band in the EELS map related to the O K-edge, indicating that O is present all along the Si/C interface of NCD-1.
The situation is different in the sample NCD-2 shown in Fig. 7. There, the EELS map of the Si/C interface indicates the limited presence of O, which is visible only in two localized green spots in Fig. 7c. This result opposes the TEM image in Fig. 7a which displays a layer of continuous bright contrast between Si and C. This case is even more pronounced in the location near and below a NiSi2 pyramid in Fig. 7b and d where the O K signal is missing at all.
Fig. 7 (a and b) HAADF STEM images of two interfacial areas between the Si substrate and the carbon layer in NCD-2; (c and d) superposition of EELS maps of K-edges of Si (blue), C (red), O (green), and Ni L-edge (yellow). The white dashed line in (d) marks the position of the profile shown in Fig. 8b. |
Fig. 8a shows the EELS spectral intensity profiles using data of individual maps of C, Si and O from sample NCD-1 plotted versus distance along the dashed line marked in Fig. 6c. The C signal (red line) sharply vanishes at the distance where the O signal (green line) reaches the maximum following the shoulder of the increasing Si signal. This quantitative figure suggests that the stoichiometry in the centre of the interfacial oxide layer rather corresponds to SiO (50% Si:50% O) than to SiO2. In any case, it is concluded that an oxide layer is preserved in sample NCD-1.
Fig. 8 EELS spectral intensities of C, Si and O (a) for NCD-1 along the dashed line in Fig. 6c and (b) for NCD-2 along the dashed line in Fig. 7d. |
In the case of NCD-2 the EELS spectral intensity profiles (Fig. 8b) of C, Si and O plotted versus distance along the dashed line in Fig. 7d shows that O is almost absent in the centre of the interfacial layer, while a small O signal (∼5%) was detected in the entire area constituted by Si indicating some surface oxidation of the FIB lamella. At the Si/C interface, the C signal (red line) drops to about 30% and then stays nearly at a constant value. The C signal almost complements the Si signal which indicates the interdiffusion of C and Si. More importantly, in the centre of the interface, the Si and C intensities are approximately equal, indicating the presence of a SiC phase. Together with the HRTEM image contrast, which points to the presence of an amorphous layer between C and Si (see Fig. 5b), it is concluded that amorphous a-SiC formed.
Based on the data shown in Fig. 5a, 6c and 8a, it is concluded that crystalline SiC cannot have formed in sample NCD-1 since a continuous oxide layer attached to Si is still preserved after the MWPECVD and following heat treatment (see Tables 2 and 3). Despite being defective (Fig. 3b and 6c) and rather thin (∼5 nm, Fig. 5a, 6c, and 6d), the SiO2 layer with reduced amount of O to 50% (corresponding to SiO) still seems to be a sufficiently strong barrier for the diffusion of C (see Fig. 8a). Nevertheless, such an interfacial barrier does not prevent Ni to penetrate through and react with the Si to form NiSi2 (Fig. 3 and 6). The question arises, why crystalline SiC occurs in NCD-2 at the base of the NiSi2. We suppose that two major differences between the samples NCD-1 and NCD-2 play an important role in the formation of crystalline SiC. Firstly, there are significant differences between the process conditions during the MWPECVD growth of NCD-1p versus NCD-2p (see Table 2), namely the higher plasma power, pressure and temperature and longer time during the NCD-2p thin film synthesis which may have more destructive effect on the initial SiO2 layer, forming an intermixed C/Si layer (a-SiC) instead (Fig. 8b).28–30 Since no SiC crystals have formed outside NiSi2 pyramids (Fig. 4 and 7), NiSi2 seems to catalytically trigger the reaction of C with Si, forming crystalline SiC. At the annealing temperature, this might occur by the crystallization of the amorphous a-SiC layer (for Si-rich a-SiC, crystallization was reported to already occur at a temperature of 1173 K (ref. 31)). Crystalline nuclei of SiC might form at the a-SiC/Si interface and subsequently grow into NiSi2 by a Ni-triggered catalytic reaction of the present Si and C which diffuse into the pyramids.
The observed chain of solid-state reactions between Ni–C, Ni–Si, and Si–C in sample NCD-2 is sketched in Fig. 9, where Ni contributes to transform diamond to graphite, reacts with Si to form NiSi2 and activates crystallization of a-SiC. Reduced to the steps 1 to 4, a similar model can be also applied for the reactions in the NCD-1 sample in cases where the SiO2 layer is defective.
Fig. 9 Scheme of the Ni–C, Ni–Si and Si–C solid-state reactions leading to NiSi2 and SiC formation at the initial Si/NCD interface in NCD-2. |
Fig. 10 (a) The temperature dependence of conductance G(T) of NCD-1 (blue curve) in comparison with G(T) of “NCD-0” published in ref. 10 (orange curve) plotted in semi-logarithmic scale. Both measured curves are fitted by eqn (1) (black dashed curves), whereas the steep region in the G(T) of NCD-1 is fitted by eqn (2). (b) Voltage dependence of conductance G(V) of sample NCD-1 measured at T = 4.2 K (dark blue), 262.4 K (red), and 322.3 K (green) exhibiting different conductance regimes. |
At low temperatures, the conductance of NCD-1 shows only a very moderate increase with rising temperature but, at about 220 K, the G(T) slope steepens and culminates in a maximum of G(T) at T ∼ 300 K. For comparison, the G(T) curve published in ref. 10 (orange curve) is added to Fig. 8a, which was measured for a NCD/graphite layer (here denoted as NCD-0) that is separated from the Si substrate by a 1500 nm thick SiO2 layer; such a thick oxide layer prohibited other reactions than the Ni-catalysed graphitization of diamond. At temperatures T < 200 K, the blue and orange experimental curves follow the same trend in conductance; however, while the orange curve of NCD-0 continues with the same trend for the whole temperature range, the blue curve of NCD-1 above 200 K strongly deviates from the low-temperature curve. This fact serves as evidence that the steep conductance slope seen in NCD-1 is unrelated to the electronic transport in the graphite layer itself. The voltage dependences of conductance G(V) (Fig. 10b) were measured at three constant temperatures 4.2, 262.4, and 322.3 K, which were chosen from the three distinct conductance regimes of G(T). While G(V) is constant at T = 4.2 and 322.3 K, G(V) exhibiting the ohmic behaviour, at the intermediate temperature of 262.4 K (steepest slope section of G(T)), G(V) is not constant but declines with increasing voltage of both polarities. The TEM images of the NCD-1 cross-section in Fig. 4 offers an explanation for the temperature-activated breakdown of the electrical current from the graphite percolating network through the defective SiO2 layer down to boron-doped Si substrate and NiSi2, which both are electrically well-conductive. Since the electrode system is fabricated on top of the NCD/graphite layer, the current passes the interface twice: from graphite to Si substrate and back to graphite. The measured G(T) curve for the NCD-2 sample is plotted in the semi-logarithmic scale as well (Fig. 11a). It shows a similar temperature-dependent electrical breakdown despite of the even more complex interfacial structure, where a part of the native SiO2 has been transformed into electrically insulating SiC.
Despite the conductance value of NCD-2 being a few orders of magnitude lower than that of NCD-1, the sudden G(T) increase similarly starts above 200 K (Fig. 10a). G(V) acquired from NCD-2 (Fig. 11b) demonstrates three different conduction mechanism regimes along the measured temperature range. At T = 4.2 K, G(V) behaves like a semiconductor with an energy gap, where current is activated by the electric field; the conductance is, therefore, enhanced by the increasing bias voltage. At T = 126 K, we identify the ohmic regime with a constant value of conductance. Finally, at T = 222 K, G(V) shows a bell-like shape typically understood in terms of hot carriers dissipating their energy in form of acoustic phonons. As the carriers gain kinetic energy sufficient to overcome the potential barrier across the dielectric interface, charge transport is dominated by the leakage current through the Si substrate. Nevertheless, while overcoming the interfacial barriers, the charge carriers transfer a part of their kinetic energy to lattice vibrations; the energy dissipation is seen as a decrease of G(V) with increasing electric field above a temperature of 200 K. The experimental results are fitted by models of conduction mechanisms considered for different temperature regions. At low temperature, the conductance through the network of percolating graphitic domains follows an expression for fluctuation-assisted tunnelling (FAT)32–34 mechanism of electronic transport:
(1) |
(2) |
The values of the fitting parameters for NCD-1, NCD-2 as well as for already published NCD-0 are summarized in Table 1.
Sample | G0 (S) | G1 (S) | kBTb (meV) | Ts (K) | ΔE (meV) |
---|---|---|---|---|---|
NCD-1 | 3.60 × 10−6 | 3.72 × 10−3 | 0.059 | 6.42 | 69 |
NCD-2 | 7.01 × 10−6 | 3.10 × 10−5 | 7.599 | 11.85 | 126 |
NCD-0 (ref. 10) | 2.14 × 10−6 | 2.46 × 10−3 | 0.170 | 9.48 | — |
The fitting parameters in Table 1 extracted from the electronic transport model corroborate our results from electron microscopy. The structure and compositional characteristics of two NCD samples differ due to various solid-state reactions taking place when the experimental setup is varied. The FAT term well reflects the crystal order of the graphite domains with a very small tunnelling barrier (kBTb = 0.059 meV) for NCD-1 showing a highly crystalline graphite structure (Fig. 2), whereas a 12-times larger tunnelling barrier (kBTb = 7.599 meV) is found for NCD-2 with disordered graphite domains and much smaller relative portion of graphite in the diamond layer. The value of the activation energy ΔE extracted from the fitting to eqn (2) for the thermally activated electrical breakdown yielded 69 meV for NCD-1, whereas a two times higher activation energy (126 meV) is obtained for NCD-2 with insulating SiC between the Si substrate and the carbon layer.
Sample | NCD thick. [nm] | Time [min] | Pressure [mbar] | Power [W] | Temperature [K] |
---|---|---|---|---|---|
NCD-1p | 70 | 24 | 60 | 3000 | 1050 |
NCD-2p | 250 | 60 | 100 | 4500 | 1250 |
In the next step, both samples NCD-1p and NCD-2p were covered with a 20 nm thick Ni film by using thermal evaporation. These samples were further subjected to a heat treatment to initialize the catalytic graphitization of the diamond surface and denoted as NCD-1 and NCD-2. The NCD-1 sample was annealed at 1073 K in vacuum (10−7 mbar) for ∼30 min (details in ref. 9), whereas the NCD-2 sample was annealed at 1173 K in a forming gas atmosphere (95% Ar/5% H2) in Carbolite tube furnace. Table 3 summarizes process parameters of the heat treatments.
Sample | Temperature [K] | Pressure [mbar] | Heat. time [min] |
---|---|---|---|
NCD-1 | 1073 | 10−7 | 30 |
NCD-2 | 1173 | 13 (Ar + 5% H2) | 10 |
For SEM, a Zeiss Supra 55 VP instrument equipped with a SE InLens and BSE detectors was used at 10 kV accelerating voltage.
Cross-sectional samples for (S)TEM were prepared by FIB using a FEI Quanta 3D FEG instrument. Using electron backscatter diffraction methods, the crystallographic orientation of the FIB lamella was carefully selected to run parallel to a (110) plane normal to the (001) surface of the Si substrate. Protective layers of Au (100 nm) and Pt (2 μm) were deposited onto the sample surface before FIB preparation to protect the sample surfaces from beam damage. Firstly, foils were thinned down to ∼40 nm with Ga ions at 30 kV accelerating voltage, and finally cleaned at 5 and 2 kV.
HRTEM and STEM was done by using JEOL JEM ARM200F (S)TEM operated at 200 kV equipped with Cs correctors for both objective lens and the condenser system. In the case of TEM, bright-field (BF) images were obtained. In the case of STEM, images were taken using a BF and HAADF detector. A double-tilt holder was used to orientate the beam direction (BD) along the [110] zone axis of the Si substrate bringing its (001) surface in an edge-on orientation. EELS images were acquired with a Gatan imaging filter (GIF) 965 Quantum ER DualEELS (convergence and collection semi-angle were 29 mrad and 74 mrad, respectively). The energy spread of the electron beam of 0.45 eV was determined from the full-width at half maximum of the zero-loss peak; the detector energy dispersion was set to 1.0 eV per channel. A dwell time per pixel of 50 ms was set and summed via 400 cycles for spectrum acquisition. Spectral images were acquired with a size of 127 × 27 pixels at a pixel size of 0.53 nm with an exposure time of 20 ms per pixel and 15 cycles.
Electrical transport measurements were performed in the four-probe configuration. Au electrodes were evaporated onto the samples and wire-bonded to Kyocera chip-carriers. The chip-carrier was located at the bottom end of a long rod with Si-diode temperature sensor (DT-470) and a resistive heating coil. The rod was placed into a vacuum-tight tubular chamber filled with He gas; an attached elastic balloon secured the atmospheric pressure of He in the chamber. A source-meter Keithley 2635B was supplying constant currents of 1–100 μA and collecting the voltage data. The sample temperature was controlled by immersing the tubular chamber with the sample at the bottom into the liquid He (LHe) container, utilizing the temperature gradient above the LHe level from 4.2 K to 300 K. The temperature of the sample was calibrated to the resistance of the Si-diode DT-470 supplied by constant current 1 μA while the voltage drop was measured by multimeter Keithley 2000. I–V curves at selected temperatures were acquired by sweeping bias voltage between ±0.05 V for NCD-1 and ±0.15 V for NCD-2.
This journal is © The Royal Society of Chemistry 2020 |