Evolution of tribo-induced interfacial nanostructures governing superlubricity in a-C:H and a-C:H:Si films

Hydrogenated amorphous carbon (a-C:H) is capable of providing a near-frictionless lubrication state when rubbed in dry sliding contacts. Nevertheless, the mechanisms governing superlubricity in a-C:H are still not well comprehended, mainly due to the lack of spatially resolved structural information of the buried contact surface. Here, we present structural analysis of the carbonaceous sliding interfaces at the atomic scale in two superlubricious solid lubricants, a-C:H and Si-doped a-C:H (a-C:H:Si), by probing the contact area using state-of-the-art scanning electron transmission microscopy and electron energy-loss spectroscopy. The results emphasize the diversity of superlubricity mechanisms in a-C:Hs. They suggest that the occurrence of a superlubricious state is generally dependent on the formation of interfacial nanostructures, mainly a tribolayer, by different carbon rehybridization pathways. The evolution of such anti-friction nanostructures highly depends on the contact mechanics and the counterpart material. These findings enable a more effective manipulation of superlubricity and developments of new carbon lubricants with robust lubrication properties.


Supplementary Figure 2 | Superlubricity behaviors and basic characterizations of the contact areas for the self-mated a-C:H (ACF-1) surfaces sliding at various normal loads.
(a) Friction coefficients μ quickly evolving to steady-state values of 0.008, 0.0055 and 0.001 at normal loads of 2, 5 and 10 N, respectively, at the onset of sliding contact in dry N2 atmosphere. The initial peak (average) Hertz contact pressures at loads of 2, 5 and 10 N were calculated to be 0.68 (0.46), 0.93 (0.62) and 1.17 (0.78) GPa, respectively. The sliding speed was 15 cm·s -1 . The inset shows the zoomed first 100 sliding cycles, in which a shorter running-in stage and a lower initial μ are observed at a higher contact pressure. (b) Optical images showing the wear scars and the wear tracks (barely visible to naked eyes) produced on the ball and wafer surfaces. The diameters of the wear scars were measured to be 97.7, 118.5 and 143.6 μm at normal loads of 2, 5 and 10 N, respectively. The corresponding steady-state apparent average contact pressures were calculated to be 0.27, 0.45 and 0.62 GPa, respectively. It seems that the self-mated hydrocarbon surfaces under a higher normal load (i.e., 10 N) were still rubbing at a higher contact pressure in the steady state even though the contact area is substantially enlarged as compared with the cases of lower normal loads (i.e., 2 and 5 N). According to the equation μ=S0/P (ref. 3) for solid lubricants, we can roughly estimate the interfacial shear strength S0 in the steady state. The calculated values were 2.16, 2.47 and 0.62 MPa, respectively, for normal loads of 2, 5 and 10 N. There was a noticeable reduction of S0 for the load of 10 N, implying a pressure-induced change of the bonding structure of the sliding interface (see Fig. 2e,h). (c,d) Interference images showing the sectional profiles across the wafer wear tracks generated at loads of 5 and 10 N, as marked in b. Wear depths of ~0 and ~11 nm were recorded for the both a-C:H wear tracks, respectively. (e) Comparison of Raman spectra (left side) measured from the as-grown a-C:H film, the ball wear scar center at 2 N, the ball wear scar center at 5 N, the ball wear scar edge and the scar center at 10 N, as marked in b. The fitting results corresponding to each curve are presented simultaneously (right side). Based on Raman theory for amorphous carbon (refs 4, 5), the almost invariable G-peak position PG (~1545 cm -1 ) and peak area ratio AD/AG (~1.5) illustrate the nearly unaffected bonding structures of a-C:H layers in the contact areas. Note that Raman signal could probe the subsurface deep to several hundred nanometers, therefore the detected information mainly reflected the properties of the bulk. However, the decreased photoluminescence background m/IG at 10 N, a measure index for hydrogen content in the film (ref. 4), implies the reduction of the detected signal volume of hydrogen atoms. This is expected to probably originate from the bulk lateral extending of the film layers ( Supplementary Fig. 3) and possible hydrogen release from the film surface upon sliding contact. Supplementary Fig. 2.  Fig. 10b). (a) Low-magnification TEM image showing the cross-sectional morphology of FIB lamella sliced from the ball wear scar, as marked in Supplementary Fig. 10d. It could be seen that there still remained a part of ACF-4 layer after the material transfer to the wafer side for growth of an anti-friction tribolayer (Supplementary Fig. 10d). (b) EDS-elemental distribution across the ball wear scar as marked in a. An outermost area in the remaining ACF-4 layer was found to be affected after the friction test when in view of the carbon and oxygen content. The increase of oxygen content was due to the oxygen adsorption from the ambient atmosphere by the nanoporous tribolayer.

Supplementary Note 1 | The strategy for EELS measurement at 200 kV
In some cases, carbon-based materials are very sensitive to high-energy electron irradiation-induced damage. The degree of damage is highly dependent on the material structure and the incident electron beam 6 . For instance, carbon nanotubes and graphene are usually suffered from knock-on damage 7 , namely displacing carbon atoms in the graphene structure.
Therefore, it would be best for the characterization of these atomic-scale thin carbon materials using electron microscopes conducted at low acceleration voltages, such as a threshold of about 80 kV for SWNT. For amorphous carbon, the motivation that we still choose the beam voltage of 200 kV for STEM and EELS acquisition is based on the following considerations. Firstly, the dual-aberration-corrected JEOL JEM-ARM200F STEM used in this work is designed to be a 200 kV microscope, which is capable of delivering a STEM-HAADF spatial resolution of 0.08 nm and an EELS energy resolution of 0.34 eV at 200 kV. This powerful sub-angstrom imaging capability is the core requirement for this study, namely resolving the tribo-induced interfacial nanostructures at the atomic scale. Although this microscope can also be operated at 80 kV, the STEM imaging resolution as well as the energy resolution would be reduced to a remarkable extent at this low voltage. This weakened imaging capability thus cannot guarantee the quality of the received images as good as those obtained in this work, which may fail in capturing the atomic details of the tribolayer nanostructures. Secondly, amorphous carbon is more endurable to electron-beam irradiation due to its bulk-like characteristic as compared to the two-dimensional (very thin) structure of graphene. Therefore, EELS spectra can be acquired at higher beam voltage such as 200 kV, by optimizing acquisition time (0.05-0.1 s) to largely suppress structural transformation and irradiation damage in amorphous matrix. Moreover, as shown in Supplementary Fig. 12, the shape of core-loss C-K spectrum recorded at 200 kV was almost the same as that of 80 kV. This indicates that similar EELS results could be achieved for both voltages. However, the relatively short acquisition time (0.05-0.1 s) leaded to the reduced smoothness of the acquired EELS spectra or even noisiness in some cases (notably those in Fig.   3f, Fig. 5j and Supplementary Fig. 11d) where carbon content in the tribolayer was not abundant.
Under such circumstances, we had to sacrifice, to some extent, the precision of bond fractions obtained through peak fitting. In spite of this, qualitative or semi-quantitative analysis of the interfacial nanostructures based on EELS spectra is definitely possible and reliable when a properly converged and stable fitting procedure (Supplementary Note 2) is consistently used for the whole set of data, even if the absolute values can be different in some degree. Figure 12 | Comparison of EELS carbon K-edge spectra recorded at 80 and 200 kV under a relatively short acquisition time of 0.05 s. Note that, for the purpose of clarity, the two spectra are normalized to the same intensity by referring to the peak intensity of C-K edge. Obviously, the almost overlapped C-K edge curves indicate that similar EELS results could be achieved for both voltages.

Supplementary Note 2 | Calculation technique for EELS C-K quantification and the accuracy of the derived π * and σ * bond fractions
The high-energy electrons can cause the electrons to excite from 1s core level to unoccupied states, and the energy-loss near edge structure (ELNES) is capable of reflecting the bonding features in C-K EELS spectra, namely π * and σ * electronic states for carbon atoms. Therefore, the C-K edge spectra can be regarded as the superposition of integrated intensities from several core-exciton π * and σ * peaks. To determine the individual bond fraction, the key requirement is to distinguish and resolve the corresponding π * and σ * peaks from the C-K edge, based on the consideration that the number of recorded counts is proportional to the number of π-/σ-bonded electrons. The fractions of sp 2 -/sp 3 -bonded carbon atoms are derived by referencing the ratio (R) of individual integrated area in the specific energy window to that of standard samples. To extract the ratios of peaks from experimental EELS spectra, a number of Gaussian or Lorentzian functions 8,9 as well as a particular energy window can be used during peak fitting. Various sets of fitting parameters are proposed in different methods 10 , and the choice of these parameters is ultimately a decisive factor in affecting the calculated bonds fractions. Numerous efforts in the literatures are devoted in determining the optimum combinations of these fitting parameters to achieve the smallest fluctuations of the derived sp 2 bond fractions, namely improving the precision of quantification. In this work, we follow a straightforward calculation method well established by Berger and co-workers 11 and other researchers 12,13 . In detail, following L.
Ponsonnet et al. 14 and A. J. Papworth et al. 9 (multiple-functional fitting approach), a few Gaussian peaks with specific energies, widths and heights are fitted to the π * and σ * features.
Note that the π * and σ * peaks involved in the present work are as follows: π * (C=C) at 285.5 eV, σ * (C-H) at 287 eV, σ * (C-Si) at 291.5 eV and σ * (C-C) at 292.5 eV. The R-ratio for each individual peak such as π * peak was achieved by normalizing its integrated area to the total (π * +σ * ) area integrated in the energy window of 280-310 eV. This ratio was then referenced to the standard value obtained for a 100% sp 2 -bonded sample such as HOPG, yielding the desired bonds fractions in the unknown carbon materials. To largely suppress the fluctuations in the fitting process, improvement could be obtained by constraining some fit parameters (e.g. the energy position of Gaussian peak) 10 . Therefore, during fitting in this work, the energies of each individual peak were fixed, while the widths and heights were variable parameters. However, it should be pointed out that, the peak energies were allowed to adjust manually in a small range around the nominal value by taking small chemical shift into account. Meanwhile, to reduce the complexity and instability during fitting, the deconvolution procedure was relatively simplified by reducing one Gaussian function, namely the σ * (C-C) peak. Correspondingly, the bond fraction of σ * (C-C) was achieved by calculating the balance from the total 100%. With fixed peak energies, the Gatan DigitalMicrograph software automatically optimized all these parameters until the deconvolution curve coincided with the experimental spectrum. For instance, during fitting of C-K edge of carbonaceous tribolayer formed from a-C:H film, two Gaussian peaks including π * (C=C) at 285.5 eV and σ * (C-H) at 287 eV were fitted to the experimental spectrum ( Supplementary Fig. 5f). The two energies of 285.5 and 287 eV were adjusted slightly, i.e., within a range of ±0.3 eV, until a good agreement between the sum of fitting peaks and the experimental spectrum was realized.
The calculation accuracy or fluctuation is the key concern in sp 2 /sp 3 carbon fractions determination from EELS C-K edge. Multiple factors such as deconvolution procedure (mainly plural scattering effect) 15 , core-hole lifetime broadening effect 16 , variable parameters (energies, width and height) in functional fitting approach 10 as well as the reference standard establishment (see Supplementary Note 3) can affect the quantitative results for each individual carbon bond.
Nevertheless, after suitable technical treatments and careful adjustments of fitting parameters, some researchers could achieve high accuracy by reducing the fluctuations in the EELS calculated bond fractions down to around 2%. In the present work, however, it should be pointed out that in addition to the above spectral issues, the less-than-ideal curve smoothness in some C-K edge curves (Fig. 3f, Fig. 5j and Supplementary Fig. 11d) due to the relatively short acquisition time (Supplementary Note 1) further impose an obstacle to obtain absolutely accurate quantitative characterization, even though we have carefully treated all these concerns by taking the well-documented fitting criterions into account. These varieties and instabilities deteriorate the precision of calculated bond fractions to a noticeable extent, namely with a fluctuation of around 10%. Therefore, based on this accuracy level, we prefer to sort the present EELS analysis results into semi-quantitative or a qualitative assessment. As mentioned above, a quantitative processing of the whole set of spectra consistently following a properly converged and stable fitting procedure at least allows a reliable qualitative analysis of the structural evolution in the targeted carbon materials.

Supplementary Note 3 | Suppression of orientation effect on EELS core edges under 'magic angle' measurement condition
As well recognized, ELNES measurements in crystalline materials are sensitive to orientation of the ordered local structure, namely the effect of anisotropy 17 . Even though the crystal orientation dependence is not the concern in amorphous carbon, the standard reference sample of HOPG used in this work would be affected by this anisotropic effect 9 . Theoretical studies have shown that there exist experimental conditions where the anisotropic effect can be cancelled when a particular collection semi-angle (β) is used 18 . This angle is the so-called magic angle. Values calculated from theoretical models for the magic angle could vary from 1.36 θE to 4θE 19,20 , where θE was the energy-dependent characteristic scattering semi-angle of electrons.
However, in practical application, there was big discrepancy between the experimental value and the theoretical one, which was speculated to be due to various possibilities such as contribution from other Bragg spots, nondipole transitions or channeling effects 21 . Therefore, for each practical case, one need to find the exact collection semi-angle for the magic-angle condition in each specific STEM-EELS system. Consequently, for each convergence semi-angle (α) at the