The carbonization of polyacrylonitrile-derived electrospun carbon nanofibers studied by in situ transmission electron microscopy

Cathode structures derived from carbonized electrospun polyacrylonitrile (PAN) nanofibers are a current line of development for improvement of gas diffusion electrodes for metal–air batteries and fuel cells. Diameter, surface morphology, carbon structure and chemical composition of the carbon based fibers play a crucial role for the functionality of the resulting cathodes, especially with respect to oxygen adsorption properties, electrolyte wetting and electronic conductivity. These functionalities of the carbon fibers are strongly influenced by the carbonization process. Hitherto, fibers were mostly characterized by ex situ methods, which require great effort for statistical analysis in the case of microscopy. Here, we show the morphological and structural evolution of nanofibers during their carbonization at up to 1000 °C by in situ transmission electron microscopy (TEM). Changes in fiber diameter and surface morphology of individual nanofibers were observed at 250 °C, 600 °C, 800 °C and 1000 °C in imaging mode. The structural evolution was studied by concomitant high resolution TEM and electron diffraction. The results show with comparatively little effort shrinkage of the nanofiber diameter, roughening of the surface morphology and formation of turbostratic carbon with increasing carbonization temperature at identical locations.


Introduction
Electrospinning is an efficient technique to provide 1-D nanostructured polymeric or polymer-based materials and composites. 1 Recent research identied carbonized polymer-derived bers as a promising class of materials for a wide scope of energy applications, such as in catalyst supports for direct methanol fuel cells, methanol oxidation and water splitting. [2][3][4] The catalytic activity for the oxygen reduction reaction in alkaline media makes nitrogen doped carbon materials a promising candidate for cathode structures of metal air batteries. 5 In addition polyacrylonitrile (PAN)-derived carbon bers also nd application in supercapacitors and as anode structures in lithium-ion batteries. [6][7][8] The tunability of the ber properties relevant for their functionalitywettability, porosity of ber surfaces, chemical composition and structure of the carbonized bersunderline the potential of this type of material. 9 However, a precondition for the tailoring of properties is knowledge and control over the effects of process specications during the preparation of the polymer solutions, the electrospinning process as well as the crosslinking and the carbonization steps.
Aer electrospinning the standard procedure to convert polymeric PAN-bers into carbon bers is oxidative stabilization, carbonization and graphitization. 10 The rst of these three steps is performed under air between 200 C and 300 C and the reactions involved are the cyclisation of nitrile and incorporation of oxygen as described by Goodhew et al. 11 This step is important to avoid votalilization, maximize the carbon yield and avoid the formation of hollow core bers in the subsequent carbonization step. 10,12 The subsequent carbonization step is conducted under inert gas atmosphere up to temperatures of about 1500 C. For higher temperatures, generally above 2500 C, the term graphitization is used. 10 Graphitization is usually performed for "high modulus" carbon bers, used for reinforced plastics and not discussed in this manuscript, which focuses on carbonization below 1500 C. The carbonization step involves losses of oxygen and nitrogen, still present aer crosslinking. The corresponding evaporation processes start at temperatures just above the stabilization temperature of e.g. 400 C. 13 However, most changes were reported for temperatures between 700 C and 1200 C. 12 Consequently changes in chemistry and structure of the bers, but also in their dimensions and surface morphology result from the carbonization step in this temperature range.
A wide scope of analytical techniques such as X-ray Photoelectron Spectroscopy (XPS), X-Ray Diffractometry (XRD), Raman, infrared-spectroscopy, thermogravimetry (TGA) and differential thermoanalysis (DTA) has been applied for the characterization of structure, crosslinking/stabilization and carbonization behavior of PAN-derived carbon bers. 7,[14][15][16][17][18][19][20] While most of these methods provide an overall view on the ber materials, a technique of choice in order to obtain localized informationin particular on the ber surfacesis transmission electron microscopy (TEM).
Most TEM-studies were performed to study strength structure-relationship of carbon microbers. In 1976 for example Bennett 21 did an extensive three-dimensional analysis of PAN-derived carbon bers heat treated at 1000 C, 1500 C and 2500 C using various TEM techniques such as bright and dark eld TEM, high resolution transmission electron microscopy (HRTEM) and quantitative electron diffraction analysis on transverse and longitudinal cross-sections. A more recent HRTEM investigation of copolymerized acrylonitrile/itaconic acid bers drawn from a spinneret in a coagulation bath is presented by Bai et al. 22 Here, amorphous and ordered structures were identied even without heat treatment. Furthermore, while highly oriented structures were detected in the longitudinal sections, the cross-sections showed onion like spherical ordering as well as crystallites.
Laffont et al. performed HRTEM, electron energy loss spectroscopy (EELS) and XRD on different types of commercial PANderived carbonized microbers. They report coherent turbostratic graphite with a d-spacing larger than 3.43Å. 23 This is larger than the value of 3.35Å published for graphite. 24,25 Moreover, the size of the stacks rstly remains small with L 10 # 4 nm (parallel to the graphitic planes) and L 002 # 1.3 nm (perpendicular to the graphitic planes), but increases with processing temperature from 300 C to 1000 C. 23 EELS analysis of the K-edges of carbon, nitrogen and oxygen shows that a carbon content of 99% is achieved for carbonization at 1000 C, while only little amount of oxygen (0.5-1% from initially 1-4%) remains in the bers. 23 More dramatic is I the change of nitrogen content from initially 1-12% it drops to zero aer 1000 C, aer 800 C 1-3% are reported. 23 In addition Laffont et al. also studied the s + p plasmon in the low loss region and show that the s-plasmon shis to higher energies for materials processed at higher temperature and correlate this with lower resistivity. 23 In a later work Laffont et al. used a combination of EELS and XPS to study the bonding situation especially of nitrogen. 26 In this study, they report an increase of about +7 at% during stabilization in air at 250 C and a loss of up to À15 at% during subsequent carbonization at up to 1000 C in N 2 -atmosphere. Depending on the ber treatment aer spinning signicant amounts of nitrogen [N]/[C] z 0.1 and oxygen [O]/[C] z 0.05 remain in the ber. The chemical composition and bonding situation of nitrogen in carbon bers were also studied in detail by other groups using XPS to distinguish different bonding types of nitrogen for application in batteries and fuel cells. 2,14,19 The previously described results were mostly related to microbers with diameters ranging from 5 to 12 mm. Musiol et al. compared PAN-derived nano and microbers heat treated at 1000 C, 2000 C and 2800 C, and reported higher massloss for nanobers, 61% residual mass for carbon microbers and 45% residual mass for carbon nanobers aer a heat treatment at 1000 C. 18 HRTEM and Raman spectroscopy showed graphitization for bers treated above 2000 C while the 1000 C bers still appear quite amorphous. 18 For the temperature range from 1500 C to 2800 C a development from relatively smooth to rough and ridged morphology was reported by Kurban et al. 27 In all reports temperature ranges differ and also a variety of different investigation techniques is applied. It is shown that the change of chemical composition; carbon structure and morphology of the bers strongly depend on the carbonization process as well as the treatment applied in the previous steps. 18,26,27 Common feature of most microscopy investigations is that the comparison of structural and morphological characteristics was performed based on ex situ experiments. Indeed, ex situ analyses are advantageous with respect to exibility of the bers processing conditions as they are not limited by the in situ experiment. However, comparisons of bers subject to small process variations require that the effects of the process variation exceed the scattering in characteristics of individual bers. Otherwise, large efforts by statistical analysis of many samples are required to provide signicant data for establishing correlations between processing parameters and structural as well as microstructural characteristics. An approach to circumvent this is, to investigate the evolution of ber characteristics on identical locations on individual bers during processing with in situ microscopy methods. A rst attempt was applied by Prilutsky et al. for the case of the carbonization of electrospun PAN nanobers containing carbon nanotubes in 2010. 28 However, in their study a heating stage, where the whole 3 mm grid is heated up, was used. Due to their larger heating volume such heating stages as a result feature slow response times, inaccurate temperatures, sample instabilities and strong thermal dri, thus limit the TEM image resolution. In this study we aim on studying the shrinkage of ber diameter, surface morphology as well as structural changes on the same bers in one single in situ TEM heating experiment at four subsequent temperature stages -250 C, aer oxidative stabilization, 600 C, 800 C and 1000 C on an in situ heating holder based on micromechanical systems (MEMS). 29 In contrast to the before mentioned system this in situ holder allows high resolution TEM images and electron diffraction in accurate temperatures without instabilities, which will help to acquire a conclusive picture for the evolution of electrospun PAN nanobers during carbonization.

Materials preparation
PAN nanobers were prepared from a dimethylformamide (DMF) solution >99% (Sigma Aldrich) containing 10 wt% PAN (molecular weight 150 000, Sigma Aldrich) using an IME EC-CLI (IME, Netherlands) electrospinner. The device was setup with a 0.8 mm nozzle diameter, a rotating cylindrical target (diameter ¼ 20 mm) and a nozzle-to-target distance of 160 mm. An electric eld of 15 kV was applied between nozzle and collector in a processing chamber, which was kept at 25 C and 20% relative humidity. The nozzle and the target were operated at lateral movement of 20 mm s À1 within a range of 100 mm and a rotation speed of 1500 rpm respectively. The feed rate for the polymer solution was 0.02 ml min À1 . The process was kept running for 30 min resulting in polymeric nanober mats of 100 mg with dimensions of 60 Â 100 Â 0.05 mm approximately. The polymeric nanober mats were dried in a cabinet at 200 C overnight to evaporate the remaining solvent. Oxidative stabilization and partial cross-linking was performed in air at 250 C for 4 h.

SEM characterization
Fiber mats were characterized by SEM (FEI, Quanta FEG 650) using an acceleration voltage of 2 kV. A micrograph of an oxidatively stabilized, non-carbonized ber mat is shown in Fig. S1. † Fiber diameters were measured manually with Olympus Stream Essentials Desktop 1.9.3.

TEM sample preparation
A dispersion of nanobers in their oxidative stabilized state, aer the crosslinking step at 250 C, was prepared by ultrasonic treatment of a piece of the nanober mat in pure ethanol. A droplet of the dispersion was applied to the heating chip of a MEMS based in situ heating holder (DENSsolutions) with Si 3 N 4 membrane and carbon coated windows. By means of focused ion beam holes of circular shape were previously etched into the carbon lm to provide areas without carbon support for the experiment. 30 In situ carbonization conditions The temperaturetime prole applied for the carbonization process during the in situ experiment is depicted in Fig. 1. The prole is designed for nanober characterization during four subsequent temperature stages -250 C, 600 C, 800 C and 1000 C. All bers located on the heating chip were heated simultaneously. Heating rate applied during all temperature steps was 15 C min À1 . Dwell times at the individual temperature steps were varying between 5 and 7 h according to the time required for TEM image acquisition of all nanobers. In total, the in situ experiment was carried out on six bers with three observations during each of the four different temperature steps. The temperature steps were split into several days of microscope time.
For three bers (1, 3 and 6) investigations were limited to low resolution size analysis, which is supposed to involve only minor inuence of the electron beam. Three other bers (2, 4 and 5) were subject to high resolution imaging and diffraction, which implies substantially higher dose for the high resolution and longer exposure to the beam during the alignment of the microscope for electron diffraction. The atmosphere during the in situ carbonization was imposed by the ultra-high vacuum (UHV) conditions required in the TEM. 31 Aer every heating period the sample holder was cooled down to ambient temperature by switching off the heating.

TEM experiments
In situ TEM experiments were carried out on a FEI Titan with a spherical aberration (C s ) corrector (CEOS) for the objective lens and operated at 300 kV using the negative-C s imaging technique, which provides images with high contrast and low noise. 31,32 In imaging mode the evolution of ber diameters, ber morphologies and alignment of the graphite layers was analyzed by low, intermediate and high resolution images respectively. The high resolution images were taken from lateral and fracture surfaces of the nanobers. Diffraction patterns were taken from one nanober at a camera length of L ¼ 490 mm where the {002} carbon reection is not covered by the beam stop and the second and third diffraction rings are still visible. Images with the same settings were taken two to three times every two hours during each temperature step.

Fiber diameter
Fiber diameters were measured on six different nanobers at different stages in course of the in situ experiment and the evolution of their diameters over time is plotted together with the temperature prole in Fig. 1. All selected nanobers have initial diameters in a range from 200 nm to just above 400 nm, which is typical for the processing used and in agreement with the ber diameter distribution measured from SEM micrographs of the ber mats ( Fig. S1 and S2 †). The evolution of the diameter of ber 6 along with the subsequent temperature treatments at 250 C, 600 C, 800 C and 1000 C is shown exemplary in low magnication TEM images in Fig. 2(a-d). The shrinkage behavior is representative for a nanober carbonized under the inuence of temperature and UHV, but largely unaffected by the electron beam. Very similar changes in diameters are observed for ber 1 and 3, with minimized exposure to the electron beam. The decrease in diameter over the complete cycle is in the range between 15 and 20% compared to the initial diameter. In contrast to that, the three nanobers exposed to the inuence of the electron beam for extended time spans (bers 2, 4 and 5) show larger shrinkage in diameter mounting up to 32%. This behavior is independent of the initial nanober diameters (Fig. 1). The enhanced shrinkage under the inuence of the electron beam occurred in particular during the 250 C annealing step in the microscope. At higher temperature, no signicant differences between the shrinkage rates can be identied.

Nanober morphology
The surface morphology can be studied at intermediate magnication micrographs as shown for ber 5 in Fig. 3(a-d).
In the initial state at 250 C in Fig. 3(a), the stabilized nanober appears smooth with homogeneous amorphous contrast. At 600 C, the roughness markedly increased ( Fig. 3(b)) and it appears like nanosized particles are sticking out from the ber. Also diffraction contrast arises throughout the nanober with brighter and darker regions with similar sizes around 5 nm.
With further increase of temperature to 800 C and 1000 C the  lateral size of the objects producing the roughness and the diffraction contrast, most probably turbostratic carbon, increases further to z10 nm at 800 C and even more up to 15 nm at 1000 C. However, the roughness itself does not continue to increase signicantly between 600 and 1000 C.

Carbon structure analysis by HRTEM
To evaluate the morphological changes of the nanober surface and the atomic structure in more detail HRTEM micrographs of the lateral surface of ber 5 at 250 C, 600 C, 800 C and 1000 C are shown in Fig. 4(a-d). Furthermore, Fig. 4(e-h) shows HRTEM micrographs of the tip of ber 4. For the initial state of the stabilized PAN-nanober at 250 C ( Fig. 4(a) and (e)) the HRTEM micrographs show an amorphous contrast and also the fast Fourier transforms (FFTs), shown in the insets, display only diffuse rings. At 600 C ( Fig. 4(b) and (f)), the ordering of the carbon atoms, mainly (002)-planes with d 002 z 0.35 nm, becomes visible in real space. At the lateral surface of the ber, this ordering appears preferentially parallel to the ber axis leading to higher intensity in the diffractogram perpendicular to the ber axis. At 600 C, only few planes are stacked, which corresponds to L 002 in the range of few nm. 23 Also the lateral size corresponding to L 10 of these turbostratic regions is in the range of few nm. 23 At 800 C, lateral and stacking size is around 5 nm. The number of planes stacked and their lateral size increases markedly in the last heating step from 800 C (Fig. 4(c) and (g)) to 1000 C ( Fig. 4(d) and (h)). Here, turbostratic regions become as large as 10 nm, in agreement with the diffraction contrast observed in the intermediate resolution images in Fig. 3(c and  d). This evolution can also be followed in the FFTs of the images shown in the insets. In the initial state at 250 C ( Fig. 4(a) and (e)), only two very diffuse rings are present. At 600 C (Fig. 4(b) and (f)), these rings get more dened and especially for the 002ring a texture appears for the HRTEM images recorded at the side of ber 5 ( Fig. 4(b)). The 002-ring shows markedly higher intensity perpendicular to the ber surface proving the preferential alignment of (002)-planes parallel to the surface. At 600 C (Fig. 4(b) and (f)) and 800 C (Fig. 4(c) and (g)), the former second diffuse ring splits into two rings. At 1000 C (Fig. 4(d) and (h)), these two rings are clearly distinguishable. In the FFTs of the images recorded at the lateral surface of the nanober, the inner of these two rings with smaller scattering vectors shows high intensity parallel to the ber axis and the outer ring has the maximum intensity perpendicular to the ber axis like the 002-ring. Comparing the scattering vectors and relative intensities for reections of graphite listed in Table S1, † the rst subring of the second ring can be attributed to {100}-and {101}planes, which are perpendicular to {002}-planes or form an angle of z72 with them (see also Fig. 5(a)). The {102}-, {004}-and {103}-planes contribute to the second subring with g hkl in the range from 5.5 to 6.5 nm À1 . Due to their multiplicity and orientation the {102}-and {103}-planes form a rather homogenously distributed intensity in the azimuthal range. The texture observed within the second subring arises mainly from 004, the second order reection of the (002)-planes. In the proposed structure sketched in Fig. 6(a), close to the nanober surface most of the graphitic (002)-planes are parallel to the electron beam and in diffraction condition. However, the rotation around the c-axis remains a degree of freedom and therefore the relative intensity of 002 and 004 is expected to be higher, compared to h0l-and hkl-reections, which are excited   Fig. 7, is shown. The rotation around the caxis remains a degree of freedom within the graphitic planes. for graphitic regions in h100i and h1À10i zone axis orientations only, as depicted in Fig. 5(a) and (b). The observed variation of scattered intensities with the azimuth could be reproduced by calculations using a simple model. Starting with the [0À10] zone axis diffraction pattern as depicted in Fig. 5(a) the variety of c-axis orientation within the plane perpendicular to the electron beam is introduced by a Gaussian distribution with s ¼ 60 around the azimuthal positions of the reections. Furthermore the relative intensities of h0l-and hkl-reections are articially decreased by a factor of 0.5 to represent relative intensities of the higher order reections in respect to 002. The resulting intensity for reections up to 103 (g hkl # 6.5 nm À1 ) is plotted as a function of the azimuthal angle in Fig. 5(c). An azimuthal angle of 0 and 180 corresponds to scattering parallel to the ber axis, 90 and 270 perpendicular to the ber axis. The thick dashed lines represent I 100 + I 101 (red) and I 102 + I 004 + I 103 (blue), which in rst approximation can be attributed to the inner and outer subring of the former diffuse second ring in the FFTs in Fig. 4(b-d). Furthermore, the graph represents the higher intensity parallel to the ber axis for the inner subring (red) and the higher intensity of the second subring (blue) perpendicular to the ber axis.
Enhanced ordering in the graphitic structure under the inuence of high temperatures is also visible at fracture surfaces at the tip of the nanober pieces (Fig. 4(e-h)). In contrast to the turbostratic structures formed on the lateral surfaces no preferential orientation of the turbostratic areas can be recognized in the real space image. Conrming this observation, the FFTs shown in the insets of Fig. 4(e-h) show rings with homogeneously distributed intensity in the azimuthal range as expected for randomly oriented crystallites. With increasing temperature these rings just get sharper and more dened. The interpretation of these observations is sketched in Fig. 6. As in the central part of the nanober the distance to the surface is comparable in all directions, so the ordered regions in this part of the nanober have arbitrary orientations among each other. Thus we conclude that ordering of carbon atoms in a turbostratic form applies also to the inner parts of the nanober. The 002-texture on the other hand is predominantly formed close to the nanober surface.
Carbon structure from electron diffraction Fig. 7 shows micrographs of ber 2 selected for electron diffraction in (a-c) and the corresponding diffraction patterns recorded at 250 C, 600 C and 800 C in (d-f). The position of the selected area aperture (SA) with a size of z170 nm in the intermediate image plane is displayed by a white circle. In contrast to HRTEM, in this experiment the complete thickness of the nanober inside the aperture contributes to the electron diffraction pattern as sketched in Fig. 6(b). As the aperture size does not include the complete nanober diameter, the diffraction patterns reveal a bit more the structure of the central part plus the upper and the lower surface of the nanober. This is complementary to the FFTs of the high-resolution images obtained at the ber surface on the side of the nanober, which is excluded by the aperture. The areas selected for recording the pattern were chosen close to the broken tip of the nanober to contain only small fraction of fracture edge, but also allowing a precise location of the SA at different stages of the in situ experiment in the TEM. At 250 C, three diffuse rings can be recognized in the diffraction pattern (Fig. 7(d)). At 600 C (Fig. 7(e)), mainly the rst, the 002-ring, becomes more dened and already starts to exhibit higher intensity for scattering vectors perpendicular to the ber axis. For the second ring the intensity increase at low scattering vectors z 4.6 nm À1 becomes sharper while the decay to larger scattering vectors stays rather diffuse. At 800 C (Fig. 7(f)), the texture in the 002-ring becomes more pronounced and the former second diffuse ring splits into two distinguishable rings, of which the rst one at lower scattering vectors is of higher intensity.The evolution of the diffraction patterns as described above can be followed in the background subtracted radial intensity, which was extracted from the diffraction patterns plotted in Fig. 7(g-i). An exponential background in the form of eqn (1) a 1 e Àb 1 (m 1 Àx) + a 2 e Àb 2 (m 2 Àx) + d (1) was tted to the decaying intensity of the primary beam. To accommodate the texture effects we extracted the radial intensity over an azimuthal range of AE20 parallel and perpendicular to the ber axis. This corresponds to an azimuthal range from 215 to 255 (perpendicular) and from 305 to 345 (parallel) in the diffraction patterns in Fig. 7(d-f). 0 and 360 represent the 12 o'clock position and the azimuth increases clockwise. The radial intensities are plotted in Fig. 7(g-i). In all plots arrows indicate the positions of reections of crystalline graphite and their vertical positions represent the relative intensities of these reections as listed in Table S1. † Within the diffraction patterns, only the rst ring originates from a single reection, which is 002. The position of its maximum at 800 C is found at z2.84 nm À1 corresponding to a d 002 -spacing of 352 pm. The second ring already contains ve reections, 100, 101, 102, 004 and 103. The sharp increase of intensity at low scattering vectors can be attributed to the rather closely packed 100 (4.694 nm À1 I rel z 3%) and 101 (4.925 nm À1 I rel z 18%). Among the ve reections contributing the second ring 101 is by far the most intense reection. The diffuse decay to larger scattering vectors can be attributed to more broadly distributed 102 (5.560 nm À1 I rel ¼ 3%), 004 (5.961 nm À1 I rel ¼ 7%) and 103 (6.482 nm À1 I rel ¼ 5%) reections. At 800 C, two separate rings become distinguishable in the formerly second ring, which can be attributed to these previously described two groups of reections. We attribute the rst subring to 001 and 101, the second subring to 102, 004 and 103 reections. As the rst subring is dominated by 101, it shows higher intensity parallel to the ber axis. For the second subring the highest intensity is observed for scattering perpendicular to the ber axis, which can only arise from 004. Both observations can be explained by the calculated azimuthal intensity distribution shown in Fig. 6(c). The same texture also inuences the relative intensity of the reections under the third main ring in the range from z7.5 nm À1 to z10 nm À1 . In this range 110 (5.4%) and 112 (8.7%) are the most intense reections and show higher intensity parallel to the ber axis.
The texture evolution can be followed by comparing the background subtracted radial intensities parallel and perpendicular to the ber axis. At 250 C, no signicant difference between scattering parallel and perpendicular (b) to the ber axis is noticed. At 600 C, the 002-intensity is already much stronger perpendicular to the ber and lower parallel to the ber axis. Perpendicular to the ber axis for the second ring a hump between 5.5 and 6 nm À1 can be recognized next to the reduced main peak, which is formed by the 101 reection (4.925 nm À1 I rel z 18%) with little contributions of 100 (4.694 nm À1 I rel z 3%). This hump can be explained with the 004 peak (5.952 nm À1 ), which is the second order reection of 002. Parallel to the ber axis 004 is less intense and the main peak is dominated by 101 and shows only long decay towards larger scattering vectors. At 800 C, this behavior becomes even more pronounced. Nevertheless, the texture effect in the two subrings of the second ring in the diffraction patterns in Fig. 7(e) and (f) is much weaker than in the FFTs of the HRTEM micrographs recorded at the side surface of ber 5 in Fig. 4(b-d).
During the last heating step up to 1000 C, the nanober selected for diffraction moved on the heating chip, so the orientation of the ber and the position of the selected area aperture are not identical to the previous measurements at the other temperatures. Nevertheless, a diffraction pattern was recorded and is displayed in Fig. 8 together with the corresponding micrograph as well as with the radial intensity extracted parallel and perpendicular to the ber axis. According to the different ber orientation, the angular ranges for extracting the radial intensities parallel and perpendicular to the ber axis changed to 175-215 (parallel) and 265-305 (perpendicular). The texture effects at 1000 C are more pronounced compared to the results at lower temperatures, with almost equal intensity of 002 and 100 + 101 parallel to the ber axis. The presence of texture in diffraction arising from a volume close to the center of the ber means the texture is also present inside the ber. The observation of diffraction rings up to 10 nm À1 correlates with an extended graphitic order. This is supported by Raman spectra measured ex situ on ber mats carbonized for 10 h in argon atmosphere at the same temperatures shown in Fig. S3. † The spectra are scaled to the D-band maximum at around 1350 cm À1 . An increase of G-band intensity around 1580 cm À1 with increasing carbonization temperature is clearly visible.

Discussion
In situ TEM carbonization experiments provide an approach to investigate the morphological and structural changes during carbonization based on observations of individual nanobers. In combination with spectroscopic methods such as EELS and energy dispersive X-ray spectroscopy (EDX) available in the TEM also the evolution of the chemical composition can be investigated in situ. By tracking the evolution of ber diameter and structure of the same nanober up to temperatures of 1000 C, statistical scattering, which stems from comparing different bers in ex situ experiments, can be avoided. In fact, analysis of SEM images (Fig. S1 and S2 †) resulted in a broad distribution of ber diameters. In situ TEM enables to measure the shrinkage in diameter on the identical ber at any time during an experiment. Aer 1000 C, we observed shrinkage in the range of 15 to 20% in diameter, which correlates to a loss in cross sectional area in the range from 27 to 35%. Musiol et al. reported a weight loss of z39% for carbon microbers and 55% for carbon nanobers measured by thermogravimetry. 18 Translating the shrinkage observed in our study into a mass loss, the shrinkage in our experiment is comparable. Differences could arise from different bers used at the initial state and the formation of porosity, which could increase the mass loss with respect to shrinkage of ber diameter.
By intermediate TEM magnication micrographs, we observed a marked increase of the nanober roughness, especially in the rst heating step to 600 C. The surface roughening was accompanied by diffraction contrast arising inside the bers. Roughening of carbon nanobers derived from electrospun PAN upon graphitization at higher temperatures of 1500 and 2800 C was also reported by Kurban et al. 27 Bennett used diffraction contrast in dark-eld TEM to image regions of turbostratic graphite in carbon microbers treated at temperatures. 21 In our study, we observe regions with sizes around 5 nm aer 600 C and up to 10 nm aer 800 C or even up to15 nm at 1000 C. These sizes agree with our HRTEM micrographs, where we estimate the size of ordered regions within the graphitic plane (L 10 or L a ) and perpendicular to them (L 002 or L c ) of few nm at 600 C, up to 5 nm at 800 C and up to 10 nm at 1000 C. So the interpretation as diffraction contrast arising from turbostratic carbon is valid. A slightly higher measure in diffraction contrast is not surprising as the eld of view and the statistics are better the image resolution poorer. In our case the estimated sizes in and out of plane are comparable. In agreement with literature we observed a preferential alignment of the graphitic planes with the ber axis. 18,21,22,27,34 This texture is conrmed by the FFTs of the HRTEM images as well as the electron diffraction patterns. Comparing our HRTEM micrographs to others the graphitic regions in our experiment appear more disordered with planes more bend and also the borders of the ordered regions are less dened. 18,21,22,27,34 Most similarity is found with the micrographs published by Kim et al. However clearly in our micrographs a higher number of planes stacked corresponding to higher L c is visible. 20 The curvature of graphitic planes observed in HRTEM is well reected by the rather broad peaks in the azimuthal distribution within the diffraction rings in the FFTs and the electron diffraction patterns. We achieved good agreement of the experimentally observed azimuthal intensity distribution with calculated intensities based on a simple model, which correlates to a variation in orientation of the graphitic (002)planes. A different degree of texture for HRTEM-micrographs recorded at the ber surface, the broken tip of the nanober and electron diffraction was observed. This can be explained with the volume probed in each experiment. In the HRTEM micrographs of the ber surface in Fig. 4(a-d) only the structure of the surface up to a depth of z30 nm is probed. With the texture proposed in Fig. 5(a), similar to that suggested by Bennet for microbers, most of the (002)-planes are parallel to the electron beam in this region of the ber resulting in enhancement of 002-reection compared to the averaged structure. 11 In the electron diffraction experiment on the other hand, the selected area aperture is positioned more central to the ber axis ( Fig. 5(b) and 7(a-c)) so diffraction arises from the central volume of the ber plus the upper and lower surface. In this part of the ber (002)-planes are expected in perpendicular orientation to the electron beam and therefore not in diffraction condition. The aperture with a diameter of z170 mm in the intermediate image plane cuts of regions close to the surface which contribute to the HRTEM-micrographs taken at the ber surface. Nevertheless texture effects are observed, underlining that, even with some distance to the nanober surface, the texture still is present. However, the HRTEM-micrographs recorded at the broken tip of a nanober don't show texture. Within the region chosen there, the turbostratic carbon is randomly oriented. We propose to apply both HRTEM and diffraction as these two techniques are complementary in respect to the volume probed. While for the analysis of nano-bers designed for mechanical applications the knowledge about the structures in the ber cores is crucial, properties of the surfaces and surface near regions are the most relevant for the application of nanobers in electrodes for metal-air batteries, fuel cells or electrocatalysis.
Quantitative analysis of the diffraction data showed that the 002 peak stabilizes around 2.84 nm À1 at 800 C and 1000 C corresponding to a d 002 -spacing of 352 pm compared to 335 pm for graphite. 25 Laffont et al. and Kim et al. reported during stabilization like our ber mat, aer 1000 C heat treatment part of the nitrogen remains in the ber. 20,26 However, in the case of an in situ TEM experiment an inuence of UHV compared to inert gas atmosphere in an ex situ experiment on d 002 -values of 349 pm and 357 pm respectively for PAN derived carbon bers. 23,34 Our result is comparable to both of them. Kim et al. attribute the stacking distance to the presence of quaternary nitrogen. 20 Laffont et al. showed that in ber set 2, which was allowed to shrink the nitrogen content is possible. If the microscope is equipped with a spectrometer, the chemical composition can be tracked by EELS or EDX during the in situ experiment.
As shown above the techniques available in a TEM allow a broad spectrum of analysis to study the carbonization of PANderived nanober during an in situ experiment. Results are at least comparable to ex situ experiments. However, it has to be kept in mind that a complete reproduction of ex situ experiments inside the TEM is not possible as parameters such as heating rate and atmosphere affect the nal properties of the nanobers. 34,36 Combined in situ heating experiments under gas ow, which are possible nowadays, could help to circumvent this issue. During the rst 250 C heating step some inuence of the electron beam was evidenced from the substantial difference in ber diameter shrinkage, depending on the exposure of the nanober to the electron beam. The latter might be attributed to the interaction of the electron beam with residual gases in the microscope and the applied heating of the sample. [37][38][39] The bers with higher exposure to the electron beam under UHV conditions at 250 C show faster shrinkage. Also an inuence on the surface roughness is possible. This could be controlled by a comparison with additional ex situ experiments on identical locations. In any case, the beam exposure should be kept to a minimum, which we did for the rest of the bers. With respect to the main subject of researchthe formation of graphitic structures in the bers depending on carbonization temperaturethis connement does not preclude the in situ method for these objectives.
With the combination of TEM at low, intermediate and high magnication and electron diffraction we were able to follow the shrinkage of ber diameter and the development of the surface morphology as well as the carbon structure during the in situ experiment. In combination with complementary ex situ experiments, the in situ experiments can contribute to an overall picture of the carbonization process for polymer-derived carbon nanobers. Thus, the in situ carbonization TEM technique provides a promising approach which is currently just at the beginning of its development.

Conclusions
Dimensional changes, surface morphology as well as the structure of PAN-derived carbon nanobers were investigated by in situ TEM during their carbonization up to 1000 C. Shrinkage in diameter as observed on individual nanobers over the whole temperature range mounts up to 20% reduction of the initial size. No marked inuence of the initial nanober diameter on its relative reduction was detected. Enhanced shrinkagein particular at the 250 C temperature stagewas indicated for nanobers, which were subject to intensive exposure to the electron beam. As a second major result, intermediate resolution TEM imaging clearly revealed the roughening of the surface, which is benecial for catalytic applications. Moreover, intermediate TEM also revealed the transition from amorphous contrast to an increased diffraction contrast on the length scale of about 5 nm at 600 C to z10 nm at 800 C and even up to 15 nm at 1000 C. HRTEM showed a similar behavior with slightly smaller sizes. Furthermore, HRTEM and its FFT showed the presence of turbostratic ordered regions for temperatures 600 C and above, with preferential alignment of the (002)-planes parallel to the ber axis on the lateral surfaces. The more graphitic structure should increase the electronic conductivity of the nanobers for electrochemical applications. Finally, electron diffraction showed that the same texture continues inside the nanober. The observation of higher order reections conrms the growth of the ordered regions. Based on quantitative evaluation of the 002-ring in the electron diffraction patterns a d 002 spacing of z352 pm was determined at the end of the experiment. Overall, in situ TEM, with its possibility of different imaging techniques, diffraction and also spectroscopic methods provides a powerful tool to study the carbonization of PAN-derived nanobers on identical locations at any time during the experiment.

Conflicts of interest
There are no conicts to declare.