Electron induced nanoscale engineering of rutile TiO2 surfaces

Electron stimulated modifications of the rutile TiO2(110) surface have been investigated using scanning tunnelling microscopy tip pulses and electron beam irradiation. Tip pulses on the ‘as-prepared’ surface induce local surface reconstruction and removal of surface hydroxyls in a region around the reconstruction. A defocused beam from an electron gun as well as tip pulses have been used to generate a number of oxygen deficient surfaces. All tip pulse features display an oval profile, which can be attributed to the anisotropic conductivity of the TiO2(110) surface. A novel oxygen deficient phase with well-ordered defective ‘nano-cracks’ has been identified, which can be produced by either electron beam irradiation or low flash anneal temperatures (∼570 K). Annealing such surfaces to moderate temperatures (∼850 K) leads to mixed 1 × 1 and 1 × 2 surfaces, until now only achievable by annealing in oxygen or ageing by repeated sputter/anneal cycles. Heating to normal preparation temperatures (1000 K) reforms the clean, well-ordered 1 × 1 surface termination. Our results demonstrate the potential of electron induced processes to modify the oxygen composition and structure of the TiO2(110) surface in a controllable and reversible way for selective surface patterning and surface reactivity modification.


Introduction
The wide-scale implementation of nanotechnology relies on the development of highly controllable and reproducible procedures to fabricate nanoscale objects. As part of this activity, patterning dielectric oxides on the nanoscale with electron beams has been demonstrated [1][2][3] along with its use to fabricate nanowires [4]. Rutile TiO 2 (110) is a model system for such work because the surface physics and chemistry have been widely studied in connection with its multiple applications [5]. Moreover, previous work has shown that TiO 2 (110) is a promising candidate for selective functionalisation of surface defects, such as oxygen vacancies (O vac ) and step edges [6,7]. In this work we show that these features can be synthesised and modified using electron induced processes.
Here we employ scanning tunnelling microscopy (STM) as a manipulation tool as well as an atomic scale imaging device. STM tip manipulation has previously been demonstrated to induce local structural alterations and reconstructions, as well as removal of adsorbates [8][9][10][11]. For example, removal of hydrogen from a H-passivated Si surface can be used to selectively create Si dangling bonds; these are then reacted with phosphine gas to dope the near surface in a highly controlled way for surface device fabrication [3,10]. STM-induced modification of structure as well as adsorbate removal have also been demonstrated for TiO 2 (110) [1,[12][13][14][15][16]. This includes the use of high current scans to remove reconstructed 1×2 areas to reveal the underlying 1×1 periodicity [12], the use of high bias scans to generate structures grown on the 1×1 surface [13], and the use of tip pulses to modify the morphology of the 1×2 reconstructed surface [14]. The removal of formate and hydrogen has also been demonstrated [15][16][17]. STM tip pulses have also been used to create areas of the 1×2 phase with a width of about 8 nm [1].
As noted above, electron beams delivered from a filament have long been used to modify surfaces, for instance by electron stimulated desorption of oxygen to form O vac in a metal oxide [2,3,17,18]. Previously, we have used this method to enhance the O vac concentration on TiO 2 (110) as well as to form (1×n) reconstructions [1,19].
In this study we combine the use of electron beams from an STM tip and an electron gun to modify TiO 2 (110). Metastable phases were created by 3 keV electron bombardment. These surfaces were subjected to electrical pulses from the STM tip, which created areas of the 1×2 reconstructed surface up to ∼160×125 nm 2 in size.

Experimental details
The measurements reported here were recorded using an Omicron ultrahigh vacuum (UHV) variable temperature scanning tunnelling microscope (VTSTM), with a base pressure of 1×10 -10 mbar. All STM images were recorded at room temperature using electrochemically-etched tungsten tips. Tips were conditioned by 10 V scans and tip pulses.
Electron beam (e-beam) irradiation was performed using an electron gun (Thermo-Electron LEG 62) defocussed to provide a spot size of 3 mm on the sample. The source was thoroughly degassed prior to e-beam irradiation experiments, and no pressure rise was detected during electron beam exposures. The beam size and position was monitored using a phosphor-coated sample plate.
Electrical tip pulses were performed in the following manner: (1) the bias voltage and tunnelling current are selected; (2) the raster scanning is paused and the tip moved to the chosen lateral position for the pulse; (3) the feedback loop is switched off and the bias voltage is changed to the selected pulse voltage and held for a chosen time period, after which the bias voltage is returned to the previous set point and (4) the raster scan is re-initiated for imaging. STM imaging and pulse conditions are reported with imaging sample bias voltage V im , imaging current setpoint I im , sample pulse voltage V p , and pulse duration t p . All imaging and pulses were performed using positive sample bias.
We note that the energy reported for electrical tip pulses is relative to the Fermi level, whereas for the e-beam irradiation the energy reported is relative to the vacuum level of the e-gun filament.

Results and discussion
As-prepared surfaces Figure 1 shows STM images of TiO 2 (110)−1×1 and the 1×2 surface reconstruction as well as corresponding structural models [5,6,[23][24][25]. The image of the 1×1 surface is shown in figure 1(a), with the corresponding model in figure 1(c). It is characterised by alternating bright 5-fold coordinated Ti atoms (Ti 5f ) and dark bridging oxygen (O br ) direction. Blue and red balls correspond to O and Ti atoms respectively, with lighter colours indicating atoms protruding further from the surface. The surface unit cell of each reconstruction is highlighted (orange boxes). The 1×2 crosslinked surface model is that proposed in [20]. rows, with bright O vac and bridging hydroxyls (OH br ) on O br rows (OH br are brighter). The image of the 1×2 surface and its model is shown in figures 1(b) and (d), respectively. 1×2 reconstructed terraces are added material on the 1×1 termination [23,25], and as a result are separated by 1×1 terminated steps (see figure 1(b)). Surface steps on the 1×1 and 1×2 surfaces are aligned in the [001], 111 [ ] and 111 [ ] directions, the detailed atomic structures of these steps being well characterised [6]. The STM features associated with the 1×1 and 1×2 terminations as well as steps form the key structural motifs that are found on the modified surfaces.
Surface tip pulses of the 1×1 termination Figure 2(a) shows the surface before the tip pulse. This serves to identify the presence of subsurface impurites in the image [26]. Figure 2(b) shows the same area after a 10 V tip pulse, with the subsurface impurities unchanged. There are several features to note. One is that H atoms on the OH br have been removed from the region around the pulse. Furthermore, after checking the effect of pulses applied from numerous STM tips using several different pulse parameters above 5 V, we conclude that the area affected is always oval in shape with the elongation in the [001] direction. This corresponds to the direction of highest electrical conductivity [27].
A number of reconstructions are observed in the immediate pulse centres. A few selected pulse induced features are shown in figure 2. We have observed a range of additional structures with sizes of ∼10-16 nm. These include high local concentrations of O vacs , 1×2 areas with crosslinking, rosette-like structures (seen previously for reoxidation of reduced surfaces [28]), and added 1×1 terraces. These reconstructions and added material are all well known phases of TiO 2 reconstructions, each with differing oxygen composition. This suggests that material is removed from and redeposited onto the surface during a pulse, as suggested in earlier work [1].
. OH br , O vac and subsurface impurities [26] (yellow circles) are present. The surface around the pulse appears 'healed' with most OH br defects removed (purple line boundary). Subsurface impurities are unchanged in location or appearance. (c) 27×25 nm 2 STM image following another 10 V tip pulse in a different area (V im =1.3 V, I im =0.02 nA, V p =10 V, t p =1 ms). The 1×1 surface rows are clearly visible. The pulse induced reconstruction consists of 1×2 rows terminating in bright features (1×2 periodicity highlighted with green lines) and with crosslinks (green circle). (d) 25×24 nm 2 STM image following a third 10 V tip pulse in a different area (V im =1.5 V, I im =0.5 nA, V p =10 V, t p =1 ms). The pulse induced reconstruction consists of 1×2 rows (green lines), a high concentration of O vac defects (blue line boundary) including elements of a c(4×2) reconstruction (black box). The region between the purple and blue lines is 'healed' by removal of OH br defects. (e) 16×17 nm 2 high contrast zoom of the pulse induced reconstruction of (b). Added 1×1 terraces (white lines indicate the 1×1 periodicity) and rosette structures (blue boxes). Orange crosses identify positions of tip pulses.

Electron gun irradiation of the 1×1 termination
TiO 2 (110)1×1 surfaces were exposed to defocused 3 keV electron beams. STM images were then recorded in three areas which we denote as: (i) the inner beam region, with a 2 mm diameter around the beam centre (highest intensity); (ii) the outer beam region, near the outer edge of the beam, 3-5 mm diameter around the beam centre (lower intensity); and (iii) outside the beam. The positions were estimated by means of the e-beam illuminated spot profile on the phosphor coated sample plate, and with distances estimated with the aid of a video camera. The beam profile is assumed to be Gaussian with higher intensity in the centre of the beam, and the electron density reported is for the entire beam. Electron current densities of 0.6-0.7 mA cm −2 have been employed for durations of 10-40 min. All images reported here are recorded after an exposure of either 0.68 mA cm −2 for 10 min (e-exp 1) or 0.65 mA cm −2 for 40 min (e-exp 2). For the longer exposure times the observed effects are amplified, but are qualitatively comparable to the results for shorter exposures. It was observed that outside the beam area the surface remained clean and the defect density remained constant. However, the inner and outer regions of e-beam exposure displayed surface modifications, as discussed below.  STM tip pulses of up to 10 V were performed on the inner beam region. Figure 4 shows the effect of such tip pulses. Large-scale reconstructions of the surface are induced with tip pulses above 5 V. Multiply-stepped 1×2 areas are observed with crosslinking, and regions of 1×1 reconstruction also form at the intersection of 1×2 steps. Figure 4(a) shows an image following a 7 V pulse. There is a ∼54×44 nm 2 oval reconstructed region elongated in the [001] direction. This oval shape is also exhibited in the overall step orientations in regions with multiple steps. As observed for pulses on the 1×1 surface, the size of the pulse-induced reconstruction is larger when the bias of the pulse is higher. Figure 4(b) shows an image taken after a 10 V pulse where a larger oval reconstructed area is observed that is ∼160×125 nm 2 in size. The 1×1, 1×2 and crosslink structures can be clearly observed in figure 4(c).
The e-beam exposed surface was also annealed to investigate the effect of heating on the transformation of the surface and to gain insight into the oxygen composition by observing the surface structures formed. The crystal was annealed to ∼850 K for 5 min. Typical STM images recorded from the inner beam region of e-exp 2 after this anneal are shown in figure 5. While the surface is still rough, the e-beam damaged surface has clearly reconstructed: small terraces have formed with 1×1 and dispersed 1×2 reconstructions. The 1×2 regions display crosslinking, whilst the 1×1 surface contains individual O vac or OH br point defects ( figure 5(b)). Further annealing to normal preparation temperatures (∼1000 K) completely lifts any e-beam induced reconstruction and returns the surface to the 1×1 phase. [ ] Therefore, it is likely that the structures are atomically ordered. Figure 6(b) shows enlarged images of some of the various atomic scale structures observed on these surfaces, including 'zipper' type [001] oriented cracks (in the green box), 111 [ ] type single (in the purple box) and double (in the blue box) line cracks, and staggered cracks composed of both types. Although the surface displays 1×1-like order, with [001] direction rows, the contrast does not match that expected for normal imaging of as-prepared surfaces. In  bright features on the terraces are indeed OH br defects, then the [001] cracks will be centred on the O br rows of the surface. However, to elucidate the exact structures of the crack structures, more detailed information is needed.
Similar structures were prepared on clean 1×1 surfaces after preparation cycles of sputtering and annealing to ∼1000 K followed by flash annealing the sample to ∼570 K without further sputtering. The sample temperature was estimated by measuring the temperature of the sample plate in contact with the sample using an optical pyrometer. The reported flash temperature of 570 K is expected to be within ±100 K of the actual temperature of the sample surface. Figure 7 shows a surface prepared in such a way. Surface cracks like those observed in the outer e-beam region in figure 6 are seen. Figure 7(a)    [ ] type single crack is clearly distinguished in the expanded green box in figure 7(b). Figure 8 displays a tip pulse at 10 V on the outer beam region. Tip pulses create a 1×2 reconstruction at the centre of the pulse as it does on the clean 1×1 surface. The pulse was performed near a step and the induced reconstruction is observed to incorporate both of the terraces. Additionally, it is observed that the 'cracked' surface around the pulse 'heals' to   recover the 1×1 order of the clean surface. Beyond the healed region, the 'nano-cracks' remain unaltered. The profile of both the centre 1×2 reconstruction and the healed 1×1 region is oval, with elongation in the [001] direction. Figure 8(b) displays the 1×1 and 1×2 periodicity, as well as the crosslinking on the 1×2 region.
The outer beam region was also imaged after annealing the crystal to ∼850 K for 5 min. The images shown in figure 5 and figure 9 were recorded on the same crystal after a single anneal. Typical STM images recorded from the outer beam region after this anneal are shown in figure 9. Large terraces with layered 1×1 and 1×2 reconstructions have formed, with the appearance that the 1×2 reconstruction is buried within the 1×1 terraces. Figures 9(a) and (b) display different imaging modes that highlight different aspects of the surface. Figure 9(a) is the 'normal' imaging mode, where individual O vac or OH br point defects can be imaged, and highlights the presence of these point defects on the 1×1 surface regions. Figure 9(b) is an imaging mode that highlights better the periodicity of the surface terraces with higher contrast. Figure 9(c) shows zoomed images that help to identify the point defects on the 1×1 surface (blue box) and single links and crosslinks on the 1×2 regions (green and purple boxes). Further annealing to normal preparation temperatures (∼1000 K) also lifts the outer region e-beam induced reconstruction and returns the surface to the 1×1 phase.
We move on to discuss the tip induced effects observed on the clean TiO 2 (110)1×1 surface. Then we discuss the insight that tip pulse induced reconstructions give into the oxygen content of the e-beam exposed surfaces. Finally, we discuss the detail of the surface modification in the outer e-beam irradiated surface, and its comparison with the structures prepared by heating the crystal.
Tip induced modifications on the rutile TiO 2 (110) surface have been reported for various tunnelling and tip pulse conditions, and lead to a variety of differing structures. Previously we have reported the formation of 1×2 reconstructions on the TiO 2 (110)1×1 surface, with domain sizes of around 6-8 nm at pulse settings of 5-10 V and tunnelling currents of 0.2-0.5 nA [1]. The latter are higher than those used in the present study (0.02 nA). We observed that the 1×1 to 1×2 reconstruction is also reversible to some extent by consecutive pulses in close proximity. This suggests a transfer of material on the surface during the tip pulse, as material is also added to the surface. Due to the large surface modification observed it was difficult to trace the movement of material precisely. At the lower tunnelling currents employed here, it is easier to identify the movement of material during the tip pulse. The central region of the tip pulse is crater-like, and added material appears around this crater. Therefore it is likely that material removed from the centre by the tip pulse is redeposited onto the surface around the crater, forming the oxygen deficient TiO 2-x structures, such as the 1×2 rows and rosette structures. The high resolution images enable us to conclude that the added features have a wellordered atomic structure. The tip pulse also clearly affects a wider area around the central region, as seen by the removal of OH br .
The oval shape that we observe in tip pulse features is likely to arise from the anisotropic conductivity of electrons deposited in the pulse, which is higher in the [001] direction as compared to the 110 [ ] direction, as reported in [27]. The mechanism for removal of oxygen is likely to be an Auger process, which is common for metal oxides. The low energy secondary electrons (<200 eV) produced by the high energy electron beam (3 keV) generate Auger transitions in the Ti-O system that causes removal of O by multiple electron loss. The neutral or positively charged O can only escape at the surface [2,30]. All of the e-beam induced effects that we observe are lifted by annealing the crystal to normal preparation temperatures. This is in contrast to the formation of 1×2 surfaces via bulk reduction, where annealing to lower temperatures does not regenerate the 1×1 surface. Instead, reoxidation of the 1×2 surface is needed to regenerate the 1×1 surface [20,28,31,32]. Hence, the e-beam induced modification of the surface affects only the near surface, and is reversible due to the ability of TiO 2 surfaces to heal themselves by exchange of oxygen from the bulk to the oxygen deficient surface (or via migration of reduced interstitial Ti species). This is in line with our previous results on e-beam exposure using lower (75-300 eV) energy electrons to generate near-surface modifications.
Further evidence that the e-beam effects are only at the near surface come from the effect of tip pulsing on the heavily oxygen deficient inner-beam region. The regions of 1×n periodicity (n2) seen after the e-beam exposure and the resulting mixed 1×1/1×2 surface after the 850 K anneal are indicative of high oxygen deficiency. When the tip pulses are performed on this e-beamed surface the large-scale reconstructions observed are surprising. They are high-energy, highly reduced surface terminations with well defined atomic-ordering, including multiple steps and 1×2 structures with crosslinking. This suggests that the oxygen deficient surface, although highly energetically unfavourable and very rough, retains substantial structural ordering within the near surface.
The modified surface structures in the outer-beam region and the anneal at ∼570 K are similar in nature. They can be explained by means of structural models based on the autocompensated step edge structures proposed in [6]. Figure 10 shows a surface model with proposed nano-crack structural features and corresponding STM images for comparison. In [6], [001] step edge structures are either terminated in O br rows (dark termination) or with nominally 4×1 reconstructed bright features, which are attributed to an 'added' TiO 2 unit where the Ti is 3-fold coordinated (Ti-3f). For autocompensation to occur, the same number of Ti-O bonds and O-Ti bonds must be broken, forming stable surface terminations. In figure 10(a) a proposed TiO 2 -unit removal is shown as a way to represent the local charge neutrality and stoichiometry compensation, allowing structural identification of removed units in the cracks. In our images a zipper-type crack is formed from two [001]-oriented steps separated by one 110 [ ] unit cell, with staggered Ti-3f on each step. Sometimes larger (Ti x O x+1 , x>1) units are observed in the crack, and periodicity not equal to 4×1 is present ( figure 10(b)). In the case of 111 [ ]-type single line cracks ( figure 10(d) figure 10(e)), the crack structure is similar to that of single line cracks, with the exception of the bright features between the dark cracks corresponding to Ti-3f or −4f as they would be at normal step edges.
It is instructive to compare features observed in the images of an e-beamed surface with those from a surface flash-annealed in UHV that are shown in figures 6 and 7, respectively. From these images we can see that single-and double-line 111 [ ]-type cracks are dominant on the e-beamed surface, whilst double-line cracks are absent on the annealed surface. Moreover, [001]-oriented cracks are dominant on the annealed surface, with longer cracks than that observed on the e-beamed surface; both surfaces are likely to have a high level of ordering of O vac . This gives insight into the relative formation energies of the features: in the annealed surface the oxygen deficient defective structures are likely to be subject to diffusion effects to generate the longer [001] cracks and highly ordered c(4×2) O vac . Also, once the cracks start to form the thermal energy from annealing will overcome the low energy barrier to crack-propagation of the stable phase step edge-like structures. In contrast, the e-beamed surface preparation involves a limited input of thermal energy. The structures are proposed to be formed via direct electronic excitation leading to bond breaking, therefore spatially constraining the formation of features to where they occur, and leading to the generation of higher energy structures such as the formation of Ti-3f in the double-line crack.

Conclusions
E-beam stimulated processes have been used to modify the surface structure of TiO 2 (110) surfaces. It has been possible to generate localised reconstruction to what are usually high temperature structures produced after reduction of the crystal in the bulk. We have also observed novel atomically ordered nano-crack structures produced with two different preparations. These processes are reversible, as annealing to preparation temperatures (∼1000-1100 K) reforms the 1×1 surface. Our results demonstrate that the oxygen composition and structure of the TiO 2 (110) surface can be modified in a controllable and reversible fashion using STM and electron beam irradiation. This provides the tools with which to create patterned structures and modified areas for subsequent nanofabrication and altered reactivity, enabling selective surface functionalisation.