Epitaxial diamond-hexagonal silicon nano-ribbon growth on (001) silicon

Silicon crystallizes in the diamond-cubic phase and shows only a weak emission at 1.1 eV. Diamond-hexagonal silicon however has an indirect bandgap at 1.5 eV and has therefore potential for application in opto-electronic devices. Here we discuss a method based on advanced silicon device processing to form diamond-hexagonal silicon nano-ribbons. With an appropriate temperature anneal applied to densify the oxide fillings between silicon fins, the lateral outward stress exerted on fins sandwiched between wide and narrow oxide windows can result in a phase transition from diamond-cubic to diamond-hexagonal Si at the base of these fins. The diamond-hexagonal slabs are generally 5–8 nm thick and can extend over the full width and length of the fins, i.e. have a nano-ribbon shape along the fins. Although hexagonal silicon is a metastable phase, once formed it is found being stable during subsequent high temperature treatments even during process steps up to 1050 ºC.

axis of the wires with 0001 oriented across the NWs. Recently nanocrystals with a dh-Si structure are grown in carbide-based film using a plasma-enhanced chemical vapor deposition method 25 .
The first transmission electron microscope (TEM) investigation of dh-Si was performed by Eremenko and Nikitenko 9 . Based on an electron diffraction study, they were able to determine the habit plane between the dc and dh phases to be {511} dc and the relationship between dc and dh Si as 110 dc ( ) // 0001 dh ( ) and [110] dc //[2110] dh . It is proposed that the {511} dc habit plane is the continuation of the coherent twin boundary between primary and secondary twins and that the phase transformation from dc to dh happened in the region where an inverse twinning shear has taken place, i.e. a twin-twin intersection or dislocation mechanism 13,[26][27][28] . In addition, Cerva found another relationship between dc and dh Si due to polytypic transformation in polycrystalline Si layers: 111 dc ( ) // 0001 dh ( ) , [011] dc //[2110] dh 29 . This relationship is also reported in some nanowires studies 18,20,21 .
In this work a new method to form dh-Si is discussed. The phase is characterized in detail with high resolution scanning transmission electron microscopy (HR-STEM) and chemical analysis. The growth procedure is based on FinFET device processing for the sub-14 nanometer technology node. Under specific conditions, dh-Si nano-ribbons can form at the base of the fins during the oxide fill step. The phase transition generally results in an outward shift of the fin which however maintains its crystal alignment relative to the substrate. Once formed, the dh-Si ribbon is stable during further processing including implantation anneal, epitaxial growth and metallization step. The phase and interface characterization and formation mechanism leading to this phase transition are discussed in detail. As the phase is in a specific epitaxial relation to the substrate and forms at well-defined positions in the devices, it opens new possibilities for integration of optoelectronic applications in standard Si-based device technologies.

Results
The important fin processing steps are schematically illustrated in Fig. 1. The TEM investigations are performed after Si fin etching (Fig. 1a), after the oxide fill and densification (Fig. 1d) and after further processing steps of the device up to first level metallization. A TEM image of groups of 12 and 4 fins after the Si etch is presented in Fig. 2a, while Fig. 2b zooms in at the group of 4 fins with higher magnification. The spacings between the fins on these images are filled with Spin On Carbon (SOC) for TEM specimen preparation. The fins are well defined with a smooth profile at the base of their sidewalls. All fins show the nominal 45 nm pitch. Due to the etch the outer fins are slightly asymmetric. The morphology of the outer fins is modified after the oxide fill and densification as illustrated in Fig. 3. In both groups of fins, the outer spacings are enlarged and the outer fins moved outwards compared to the configuration after the etch. The width of the fins is slightly reduced by the oxidation. The profile at the bottom and the pitch of the inner fins are unchanged. The outer fins appear wider than the inner fins and have a typical bulge at the outer bottom and a step on the silicon on the inner side. The direction from step to bulge is nearly 16º inclined upwards ( Fig. 3c and 4) to the outside of the group of fins, i.e. corresponds to a {115} dc plane.
Step and bulge have similar widths and indicate an outward shift of the fin along this plane (Fig. 1d). Zooming in at the bottom of the outer fins after oxidation and annealing, shows 4 possible configurations which occur in a ratio 45/15/30/10 in the over 300 studied finfet structures: a defect across the full fin width (Fig. 4a), a partial defect (Fig. 4c), no defect but bulge and step present with similar dimensions as for the defected cases (Fig. 4d), and rarely no defect and no clear bulge or step. The defects consist of a slab of material of different phase with a thickness of about 5 to 8 nm embedded between dc-Si. The full defects run from the step to a position below the bulge as illustrated in Fig. 4a. The width of the defected region is narrower than the Si below and above (including the step and bulge respectively). Figure 4b shows a high angle annular dark field (HAADF) HR-STEM image taken at 300 kV. The dumbbell structure which can be seen in the silicon substrate and fin, is also present inside the defect with similar brightness of the atomic columns, i.e. same composition but different structural arrangement of the columns. Energy dispersive X-ray spectroscopy confirms that the material consists of Si only. The partial defects are always situated next to the step at the inner side of the outer fins as illustrated in Fig. 4c.
Detailed crystal structure analysis of the slab material is shown in Fig. 5. The atomically resolved HAADF STEM image viewed along the fin direction (Fig. 5a) shows that the defect consists of a single crystalline material which is in epitaxial relationship to the dc-Si of substrate and fin. The crystal structure is analyzed from the Fourier transform (FFT) of the image (Fig. 5b). The sharp bright spots in the FFT pattern are due to the dc-Si of substrate and fin along [110] zone axis. The streaked spots with lower intensity are from the slab material. The spots are doubled and elongated due to a rotation of the lattice in the left and right part of the defect. The reflections can be indexed as shown in Fig. 5b  , which is the same epitaxial relation as observed for diamond-hexagonal silicon in indentation structures 9 . The lattice parameters a and c are determined from the FFT pattern as 0.38 nm and 0.63 nm respectively with the ratio of c/a = 1.65, i.e. very close to the ideal value for dh-Si as can be calculated based on the dc-Si parameters. The majority of the analyzed defects shows a stepped configuration as on Fig. 4a and 5a, i.e. part of the defect at the inner side of the fins show 001 dc ( ) interfaces and next part has sloped interfaces parallel An atomically resolved HAADF STEM image at the slab area of a sample cut parallel with an outer Si fin is presented in Fig. 5c. The dh-Si ribbon is magnified further as insert at the bottom right. In the dc-Si of fin and substrate the typical dumbbell pattern is observed, while in the defect region clearly the regular hexagonal pattern corresponding to the dh-Si atomic columns along [0001] dh can be seen. The  visibility of the latter pattern will only be ideal for defect parts that are observed exactly along the [0001] dh zone axis, and with the interface 001 dc ( ) // 0110 dh ( ) extending on-edge through the full TEM specimen thickness. In view of the defect dimension and the step configuration, even in a very thin sample, this condition will only rarely be fulfilled. Superposition of dc and dh lattices in the viewing direction will result in vertical streaking of the atomic columns. In the inclined defect parts with 115 dc ( ) interfaces, this streaking will be further enhanced by the 4º rotation of the dh lattice. Therefore the clear visibility of the dh-lattice is limited to smaller areas along the defect. Observations at different positions along the length of the fin show that the dh-lattice can be recognized in regions with several hundred nanometer length, i.e. nano-ribbons of dh-Si are present at the base of the fins. The dh-regions are separated by parts of the fin that appear as dc-Si, which will correspond to the 3 rd (or 4 th ) kind of configuration as observed in the across fin views, i.e. step/bulge without defect. The relationship between dc-Si and dh-Si for the observation of the cut parallel with fin, is presented by the insert at the bottom left of Fig. 5c and the FFT shown in Fig. 5d. It fully agrees with the epitaxy derived from the cut perpendicular to the fin. The combined information from both observation directions is fully conclusive for the dh-Si nature of the defect and excludes the possibility of presence of inclined twin defects as discussed for dh-Si nanowires 4,5 .
Many partial dh-Si regions show a planar defect at their ends in a {111} dc plane directed towards the silicon substrate (Fig. 6a). The lattice resolution is not always clear at these defects, but where interpretable they correspond to stacking faults. In TEM observation mode these defects are often strongly electron beam sensitive, i.e. grow during imaging. Stacking faults bonded by a 1/6[112] Shockley partial are also present at the dc/dh transition observed in samples cut parallel with the fin (Fig. 6b).

Discussion
The crucial process steps that lead to the diamond-hexagonal silicon formation are schematically illustrated in Fig. 1. After the Si etch of the fin structures, the sidewalls have a smoothly varying profile and the silicon is defect free. The fin morphology is unchanged during the oxide fill step but is modified during the subsequent wet oxygen anneal. The oxidation process is discussed in more detail on Supplementary Fig. S1 and S2 online. Major change due to the oxidation is an outward shift of the outer fins characterized by an increase of the width of the last spacing, the formation of a bulge on the outside and of a step on the silicon at the bottom of the outer spacing. In about 60% of the outer fins this goes together with the formation of a hexagonal silicon slab over the full or partial width of the fin. Most samples investigated are further cured in N 2 at higher temperature and received also additional temperature treatments during the further processing of the devices. All these subsequent steps are found to have no impact on the presence or morphology of the hexagonal silicon ribbons. Hence the key process step that causes the phase transition to hexagonal silicon is the wet oxidation while the formed dh-phase is stable during further thermal steps. The wet oxidation simultaneously densifies the deposited oxide and oxidizes the sidewalls of the fins, i.e. the fin width w decreases and the spacing s increases with the same amount. The conversion of Si to SiO 2 results in a volume expansion by a factor 2.  the oxide is not viscous and is therefore not pushed upwards out of the trench. Hence the mechanical stress is horizontally exerted on the sidewalls of the fins and is maximum near the bottom (Fig. 1c, Supplementary Fig. S1e online). The fins are slightly tapered after the etch so that the initial spacing s varies from about 25 nm at the top to 10 nm at the bottom of the fin which leads for the applied oxidation conditions to stress of about 19 GPa and 35 GPa respectively. Part of the volume expansion will however be compensated by the densification of the deposited oxide. If half of the expansion is accommodated by the densification, the stress will be reduced by nearly a factor two, and is therefore still considerable. For inner fins the inward and outward stress on both sides is quasi the same. However, for the outer fins the outward stress on the inner side of the fin is much larger than the inward stress on the outer side because the space to the next structure is in micrometer range and the estimated stress at that side remains below 1 GPa (Fig. 1c). Hence, the phase transformation and outwards shift of the outer fins can be related to the strain/stress from the unbalanced expansion during the wet oxidation step of the wide and narrow oxide windows next to the outer fins. The estimated stress levels are in the range needed to have a transition from dc-Si to dh-Si and it can be expected that the transformation will be initiated near the bottom of the fin where the stress is the largest. Once the transformation is finished and the fin shifted, the stress in the outer fin is released and no further transformation will occur during further processing steps that are not leading to additional oxide expansion. Moreover the high temperature N 2 anneal after the oxidation step will allow vertical viscous flow of the oxide so that a general stress reduction occurs. The bulges and inner steps have similar sizes for all 3 major types of configurations at the bottom of the outer fins. The occurrence of similar fin shifts also without presence of dh-Si and with partial dh-Si slabs indicates that the partial dh-regions are not the nucleation state of the full dh-slabs. A transformation of mass from the inner side to the outer side of the fin could occur by dislocation nucleation at inner side, glide through the fin width and annihilation of the dislocation at the outside. However the bulge should then be situated much higher so that the step to bulge plane would correspond to a (111) dc glide plane. Dislocation formation and glide in Si at the low densification temperatures is however very unlikely and below 700 º C, plastic deformation in Si occurs by other mechanisms such as twinning and as shown in the present study by phase transformation. Moreover, dislocation creation and glide cannot lead to dh-Si formation.
Transformation of dc-to dh-Si in indentations experiments 9,13,26,28 is explained as a stress induced martensitic transformation involving shear at the intersection of the secondary twins which are generally present in high density in such indents 13,26 . In our case, only single stacking faults are observed at the edges of the dh-regions but never extensive twinning is observed. Therefore such martensitic transformation mechanism seems less likely to explain the dh-Si formation. A model for direct transition of dc to dh-Si by application of high stress on the (110) dc plane was proposed by Tan et al. 12 . It involves bond breaking and the formation of new 6 atom rings which after relaxation result in dh-Si with the 0001 dh ( ) plane parallel 110 dc ( ) . The transformation results in a 18% volume contraction in the [110] dc direction and equal expansion along [001] dc while no volume change occurs in the [110] dc direction. Such mechanism could explain the presence and size of the steps on the inner side of the spacing and the reduced width of the dh-Si slab compared to the dc-Si above and below. To explain the shift of the fin also in cases of partial or no dh-Si, a reverse transformation mechanism should occur once the stress is released so that the dc lattice is recovered. Although dh-Si is generally considered to be a metastable phase 7 , the observed stability of the dh-Si slabs during further processing is in contradiction with such reverse transformation. An alternative process that overcomes this problem could be a dc-to dh-Si transformation that involves a cycle through several high pressure Si phases as observed in high pressure and indentation experiments. In such experiments an irreversible transition to β -tin Si (Si II) is observed for pressures above 12 GPa 10,11 which is in the estimated pressure range at the bottom of the fins. The high pressure β -tin phase shows a 30% volume reduction per Si atom compared to dc-Si which can explain the formation of the step and reduction of the lateral width of the final dh-Si slab compared to the Si above and below. Upon unloading, the β -tin phase relaxes through a sequence of Si phases. In a combined Raman and TEM study with in-situ anneal of nanoindents in Si, Ge et al. 10 showed that the phase sequence for slow reduction of the stress involves Si II (tetragonal ) → Si XII (rhombohedral) → Si III (bcc) → (Si XIII) and then coexistence of dh-Si (Si IV) and amorphous Si. At the oxidation temperature used for the densification step of the oxide fill a thin region of amorphous Si confined between the dc-crystalline substrate and fin can easily epitaxially regrow. Therefore following such phase sequence the final state could be the coexistence of dc-and dh-Si. The formation of either phase in the transition region might be favored by the local shape and interfaces of the intermediate phases. As these phases are metastable and quickly anneal out above 300 ºC 10,11 , they will not be present anymore after the oxide densification step and no direct evidence of this phase sequence is observed in our experiments. Nevertheless, such process could explain the presence of step/bulge with partial or no hexagonal silicon slab combined with stability of the hexagonal phase during further processing.
Most dh-Si slabs show a change of the interface with the dc-Si from inner to outer side of the fin. Whereas the 001 dc ( ) // 0110 dh ( ) interface corresponds to the perfect alignment of 110 dc ( ) with 0001 dh ( ) , the 115 dc ( ) to 0332 dh ( ) interface implies a rotation over 3.5 º as experimentally observed. The latter interface is discussed in literature 12,13 whereas the first case is not reported before. Both interface structures can be modelled with 5 and 7 atom membered rings without any dangling bonds (Supplementary Fig. S3 and S4 online) and show over a nearly 2 nm distance common lattice positions with only a small misfit (Table 1). However for the 001 dc ( ) // 0110 dh ( ) interface the reshuffling of the bonds needs to be spread over 2 atomic planes resulting in a stepped dc/dh interface which will be less stable over longer distances ( Supplementary Fig. S3 online). Steps at the interface may then result in a switch to the more easily formed 115 dc ( ) interface. The formation of dh-Si requires sufficiently high stress to initiate the phase transformation and therefore does not occur for conditions that lead to lower stress, e.g. by applying less severe oxidation conditions (lower temperature, shorter oxidation time) or with larger initial spacing. As the dh-Si is situated at the base of the fins, it has no direct impact on the electronic properties of the devices that are processed on top of the fins as they only depend on the upper ~50 nm part of the fins. On the other hand by controlling the oxidation conditions and relative spacings on both sides of the fins, conditions that favor the formation of a dh-Si ribbon can be optimized. As the position and crystallographic orientation of the resulting dh-Si are well controlled, this opens possibilities to combine the optical properties of this phase in opto-electronic devices with advanced FinFET based nano-electronic devices. For example, by gradually increasing the spacing between the fins one could generate a series of parallel dh-Si nano-ribbons which could be used as waveguides in the devices. Further investigation of both design and processing conditions is needed to explore the possibilities of such applications. Conditions that lead to increased dh-Si volumes (thicker, higher density of ribbons) are also needed to proof the optical functionally of the material by cathodoluminescence, the sensitivity of which turned out to be insufficient for the present material.

Methods
Bulk FinFET process. The investigated FinFET structures are taken from a range of wafers processed up to different end-steps for process and device optimization. After the Si etch of the bulkfin structures the spacings between the fins are filled with a CVD oxide (O 3 /TEOS)). To reduce the void in between the fins and to densify the deposited oxide a wet oxidation anneal at low temperature (700-750 ºC) or a combination of wet oxidation anneal and high temperature inert cure (>1000 ºC) are applied. Figure 1 presents a schematic drawing of the process steps involved relevant to this investigation. The major steps are similar as discussed by Redolfi et al. 31 but in the present work self-aligned dual patterning technology is used to obtain 14 nm node fin dimensions with fin pitch of 45 nm 32 . The investigated fin structures consist of groups of 4, 10 or 12 fins with constant pitch. Due to the etch, the sidewalls are sloped so that the spacing typically varies after etch from 25 nm at the top to 10 nm at the bottom. The groups of fins are separated by distances of more than 500 nm (Fig. 2a). The fins have lengths of 250 to 2500 nm and heights in the range of 100-130 nm.
TEM specimen preparation. All the TEM samples are prepared by Focused Ion Beam (FIB, FEI Helios450HP) with the in-situ lift-out technique. To protect the sample surface during the ion milling a Spin-On-Carbon (SOC) layer is deposited on the full wafers or wafer pieces. This cap material fills high aspect ratio topography well and weakly planarizes the surface. The region of interest is further locally capped in the FIB with electron beam or ion beam deposited Pt. The major milling is done with a 30 kV Ga ion beam while the milling progress is controlled with the scanning electron microscope. Final milling to minimize the damage layer on the specimens is performed with 5 kV Ga ion beam. The typical TEM specimen thickness is ~50 nm. Most samples are prepared across groups of 4, 10 or 12 fins. In addition also TEM specimens are prepared parallel with the fins so that only the outer fin of the groups is included in the TEM specimen.

TEM analysis. Investigation by transmission electron microscope is performed with FEI Tecnai F30
or Titan 60-300 Cube in TEM or STEM imaging mode. The high resolution STEM images are acquired in the double corrected Titan system with the high-angle annular dark field (HAADF) detector generally at 120 kV, unless specified otherwise. This condition allows high resolution that resolves the Si-dumbbells without noticeable beam damage in the specimens. The images are acquired with a screen current of 0.1-0.2 nA, a convergence angle of 22 mrad and an inner collection angle of 50 mrad. HAADF contrast, also known as Z-contrast, can provide directly not only the location of the column of atoms but also information on the local composition at atomic scale. However, resolution can be further improved by using 300 kV.