Digital Etch Technique for Forming Ultra-Scaled Germanium-Tin (Ge1−xSnx) Fin Structure

We developed a new digital etch process that allows precise etching of Germanium or Germanium-tin (Ge1−xSnx) materials. The digital etch approach consists of Ge1−xSnx oxide formation by plasma oxidation and oxide removal in diluted hydrochloric acid at room temperature. The first step is a self-limiting process, as the thickness of oxide layer grows logarithmically with the oxidation time and saturates fast. Consistent etch rates in each cycle were found on the Ge1−xSnx samples, with the surfaces remaining smooth after etch. The digital etch process parameters were tuned to achieve various etch rates. By reducing the radio frequency power to 70 W, etch rate of sub-1.2 nm was obtained on a Ge0.875Sn0.125 sample. The digital etch process was employed to fabricate the Ge1−xSnx fin structures. Extremely scaled Ge0.95Sn0.05 fins with 5 nm fin width were realized. The side walls of the Ge0.95Sn0.05 fins are smooth, and no crystal damage can be observed. This technique provides an option to realize aggressively scaled nanostructure devices based on Ge1−xSnx materials with high-precision control.

and 3.2 nm/cycle for 10 s and 60 s etch duration, respectively) 34 . In addition, the formation of SnF y is dependent on the initial Sn content, which limits its application for Ge 1−x Sn x with low Sn content.
Here we present a new digital etch process that allows the etching of Ge or Ge 1−x Sn x structures with sub-1.2 nm precision. The digital etch process is realized at room temperature, which avoid the Sn surface segregation or precipitation. The Ge 1−x Sn x surface after digital etch is smooth, and it can be used directly for extremely scaled Ge 1−x Sn x fin formation in Ge or Ge 1−x Sn x CMOS applications. In addition, such a unique process capability can also enable fabrication of Ge 1−x Sn x based nanostructures for a range of Si-compatible photonics and microelectro-mechanical (MEMS) device applications.
The Ge 1−x Sn x samples used for this study were grown on 4-inch Ge(100) substrates by a solid-source MBE system. The Ge substrates were cleaned by diluted hydrofluoric acid (DHF) before being loaded into the load-lock chamber of the MBE system. After transferring the substrates into the growth chamber, annealing was performed at 630 °C for 5 minutes to remove the native oxide. The substrate temperature was then adjusted to 100-150 °C for Ge 1−x Sn x growth. The thicknesses of the Ge 1−x Sn x layers are kept below 50 nm, so that all the Ge 1−x Sn x films are fully strained to the Ge substrates 20 . In addition, a commercial GeOI wafer was used to extract the etch rate of Ge.
To avoid the problem caused by thermal instability of Ge 1−x Sn x layer having a high Sn content [36][37][38] , a low temperature process is preferred. Our digital etch approach forms a Ge 1−x Sn x oxide through exposure to oxygen plasma in an asher, and removes the oxide in dilute hydrochloric acid HCl (10%) for 30 s at room temperature. The plasma oxidation was performed at a pressure of 300 mTorr. As the oxidation reaction (first step) is self-limiting, the etch depth is no longer dependent on the etch time, but is dependent on the number of etching cycles. By repeating these two steps, the extremely scaled Ge or Ge 1−x Sn x fins can be formed in a controlled manner as shown schematically in Fig. 1.

Ge 1−x Sn x Oxide Formation by Low Temperature Plasma Oxidation
It was reported that migration of Sn atoms from the Ge 1−x Sn x surface and from inside the Ge 1−x Sn x to the Ge 1−x Sn x oxide occurs during thermal oxidation 39 . Instead of using a conventional thermally activated oxidation at high temperature, a plasma activated oxidation is used as it can be done at low temperature. To better understand the mechanism of the digital etch process on Ge 1−x Sn x , X-ray photoelectron spectroscopy (XPS) was performed to investigate the formation of Ge 1−x Sn x oxide during plasma oxidation, and removal of the surface oxide in HCl solution. Figure 2a and b show the Ge 2p 3/2 and Sn 3d photoelectron core spectra of the Ge 0.875 Sn 0.125 /Ge sample after plasma oxidation, with radio frequency (RF) power of 250 W and plasma oxidation time of 120 s. Both the peaks of the Ge and Sn oxides can be observed in the Ge 2p 3/2 and Sn 3d photoelectron spectra, respectively. The binding energies (BEs) of these peaks agree well with the reported values of BEs for stoichiometric GeO 2 and SnO 2 39,40 . As Ge 2p 3/2 photoelectrons have a short inelastic mean free path λ (λ for Ge 2p 3/2 is ~9.7 Å) 41 , no Ge peak can be observed in Fig. 2a, while clear Sn peaks can be identified in Fig. 2b. The core-level spectra were fitted using a combination of Gaussian and Lorentzian line shapes, together with a Shirley-typed background substraction. This results in the overall blue line fitting of the core-level spectra with their respective peak components (gray lines). The Sn content in the Ge 1−x Sn x oxide layer can be calculated using A Sn-O /(A Sn-O + A Ge-O ), where A Sn-O and A Ge-O are the normalized areas of the Sn-oxide and Ge-oxide peaks, respectively. We found that the Sn content x in the Ge 1−x Sn x oxide layer is 13.8%. Figure 2c and d show the Ge 2p 3/2 and Sn 3d spectra of the Ge 0.875 Sn 0.125 /Ge sample after plasma oxidation and HCl dipping. There were no obvious peaks related to Ge or Sn oxides in the Ge 2p 3/2 and Sn 3d spectra, indicating that the Ge 1−x Sn x oxide formed by oxygen plasma can be effectively dissolved using a wet chemical treatment with dilute HCl solution. In Fig. 2c, a small shoulder can be observed besides Ge peak, which indicates a minor BE peak at 1218.9 eV after fitting. The peak is associated with native oxide on the Ge 1−x Sn x layer surface formed from atmospheric exposure before the ex situ XPS measurement. The smaller BE of the observed Ge oxide peak compared with that of stoichiometric GeO 2 indicates suboxide GeO x formation on the Ge 1−x Sn x layer owing to atmospheric exposure. The BEs of Sn 3d peaks slightly increases after HCl dip, which is attributed to the band bending originating from different charged states at the GeSn oxide/GeSn interfaces 42 . The Sn content in the surface Ge 1−x Sn x region was calculated to be 12.0% using the normalized areas of the Sn and Ge peaks (Fig. 2c,d), which is close to that in bulk region, and slightly smaller than that in the Ge 1−x Sn x oxide layer formed by plasma oxidation. This result indicates negligible Sn atoms migration from inside the Ge 1−x Sn x to surface occurs during the plasma oxidation at room temperature. Digital Etch of Ge and Ge 1−x Sn x layers. High-resolution X-ray diffraction (HRXRD) was used to investigate the Sn content and crystallinity of the Ge 1−x Sn x samples. Figure 3a shows the HRXRD scans around the 004 diffraction point for the Ge 0.875 Sn 0.125 /Ge samples before and after digital etch. Clear GeSn and Ge peaks can be identified for the as-grown Ge 0.875 Sn 0.125 sample, and the presence of interference fringes surrounding the GeSn peak suggests good interface quality. The full width at half maximum (FWHM) of the GeSn diffraction peak increases with increasing etch cycles, which indicates the decrease of Ge 0.875 Sn 0.125 layer thickness according to Scherrer's formula 43 . This is consistent with the increasing interval of the interference fringes after digital etch. After an 8-cycle digital etch, the GeSn peak moves slightly towards the Ge peak. The Sn content can be calculated to be 11.7%, which corresponds to a Sn content decrease of approximately 0.8%.
On the other hand, X-ray reflectivity (XRR) measurements have been performed on the Ge 1−x Sn x samples before and after digital etch. XRR can be used to determine thickness of thin film with high accuracy by analysing X-ray reflection intensity curves from grazing incident X-ray beam. As a non-destructive technique, no additional lithographic patterning is needed for etch depth calibration. Figure 3b shows the experimental (open black circle) and simulated (blue curve) XRR curves of the as-grown Ge 0.875 Sn 0.125 , the Ge 0.875 Sn 0.125 after plasma oxidation, and the Ge 0.875 Sn 0.125 after HCl dipping. The thickness of the as-grown Ge 0.875 Sn 0.125 film is 30.7 ± 0.2 nm, which can be obtained by fitting the reflected X-ray intensity versus incident angle. The thickness of the Ge 0.875 Sn 0.125 film decreases with increasing number of digital etch cycles. After a 4-cycle digital etch and plasma oxidation, the thickness of Ge 0.875 Sn 0.125 film drops to 19.4 ± 0.2 nm. The etch rate can be estimated to be 2.2 nm/cycle. In addition, the thickness of Ge 1−x Sn x oxide formed by plasma oxidation can be also extracted by XRR, which is 3.9 ± 0.2 nm. Based on the molar mass and density of Ge and GeO 2 , for every 2.2 nm-thick Ge consumed, 3.97 nm-thick GeO 2 will appear. This is consistent with the XRR results. Figure 4a shows the etched layer thickness as a function of etch cycle number for Ge 1−x Sn x samples with various Sn content. The RF power is fixed at 250 W, and the plasma oxidation time is fixed at 120 s. The etched thickness for the Ge 1−x Sn x samples was extracted from XRR, while that for the GeOI was obtained from ellipsometry measurements. Good linear fits can be observed in all the samples, which confirm that a consistent depth of Ge or Ge 1−x Sn x is etched away each cycle. The etch rate extracted from the slope is ~2.2, ~1.9, and ~1.4 nm/cycle for the Ge 0.875 Sn 0.125 , Ge 0.95 Sn 0.05 , and Ge, respectively (Fig. 4b). It was reported that SiGe alloys had higher oxidation rates than Si under thermal oxidation conditions 44,45 . This enhancement under thermal oxidation increases with increasing initial Ge concentration in the alloy. Therefore, it could be explained in terms of the Si−Ge bond being weaker than the Si−Si bond. Similarly, the presence of Sn atoms results in weaker Ge-Sn bonds than Ge-Ge bond, and weaker still Sn-Sn bonds at the sample surface (at T = 298 K, the dissociation energies for the Ge-Ge, Ge-Sn, and Sn-Sn bonds are 264.4 ± 6.8, 230.1 ± 13, and 187.1 ± 0.3 kJ/mol, respectively) 46 , which may lead to the increased oxidation rate (etch rate) with increasing Sn content. In addition, the introducing Sn may slightly degrade the residual order in the oxide layer, which also affect the etch rate.
Surface area to volume ratio increases with downscaling of device dimensions and surface properties can appreciably affect device electronic properties. Atomic force microscope (AFM) was employed to investigate the surface roughness of the Ge or Ge 1−x Sn x samples after digital etch.   was found that the RMS roughness of GeOI samples after etch remains as small as ~0.11 nm. Figure 5f summarizes the RMS roughness of Ge and Ge 1−x Sn x samples after various digital etch cycles. The digital etch increases the RMS roughness slightly for Ge 1−x Sn x samples but not for GeOI samples. Both GeO 2 and SnO 2 were formed on Ge 1−x Sn x samples during plasma oxidation. The slightly increased RMS roughness after successive etching cycles may be related to the residual SnO 2 after HCl dip in which SnO 2 masks subsequent oxidation, or related to the different oxidation rates for Ge and Sn. Figure 6a shows the measured Ge and Ge 0.875 Sn 0.125 digital etch rates as a function of plasma oxygen exposure time. The RF power is fixed at 250 W. It is noted that for both the Ge and Ge 0.875 Sn 0.125 samples, the etch rate increases rapidly for the first 120 s of oxidation time, and then saturates fast to a etch rate of ~1.4 and ~2.2 nm/ cycle, respectively. Deal and Grove proposed that the thermal oxidation rate could be determined by a combination of two processes 47 . One is the actual chemical reaction of oxygen O at the interface, and the other is the diffusion of oxygen through the previously formed oxide film. The combination of these processes resulted in   49 . They found that the oxide layer thickness y grows logarithmically with increasing oxidation time t as shown below: 2 where A and B are the parameters related to the effective diffusion coefficient of O atoms in the oxide, and the concentration of O atoms at the oxide surface and in the oxide layer; τ is the oxidation time for native oxide formation. This logarithmic limit shows faster saturation of the diffusion-limited process compared with the parabolic limit from the Deal-Grove model. Simply assuming τ = 0, good experimental fits for the etch rate (oxide thickness) were obtained with this theory, as shown in Fig. 6a (dash line). To achieve digital etch with higher precision, the etch rate should be reduced. As discussed, the etch rate can be tuned by varying the oxidation time. However, an ideal digital etch should eliminate the requirement f or timed etching, and decreasing oxidation time may lead to larger variation of etch rate. The etch rate can be also tuned by varying the RF power or gas pressure. Figure 6b shows the etched thickness of Ge 0.875 Sn 0.125 as a function of etch cycle, and the etch rates were extracted in Fig. 6c. The oxidation time is fixed at 120 s. The etch rate decreases with decreasing RF power, and etch rate of sub-1.2 nm/cycle was obtained with the RF power reduced to 70 W. The main oxidizing species should be atomic oxygen during plasma oxidation 49 . As the atomic oxygen concentration increases with higher RF power, the etch rate increases with higher RF power 49, 50 . Formation of Extremely Scaled Ge 1−x Sn x Fins. Digital etch technique provides high-precision control of etch, and can be used to realize extremely scale devices, such as deeply scaled FinFETs. According to the 2013 International Technology Roadmap for Semiconductors (ITRS) Overall Roadmap Technology Characteristics (ORTC) 51 , sub-5 nm fin width is required in the sub-10 nm technology nodes. Here we explore the possibility of realizing Ge 1−x Sn x fins with 5 nm fin width.
The ~28 nm-thick Ge 0.95 Sn 0.05 /Ge sample was used for forming fins. After electron beam lithography (EBL) patterning and chlorine-based dry etch, a trapezoidal Ge 0.95 Sn 0.05 fin structure was formed, with fin height of ~70 nm and sidewall slope of ~75°. Details of the Ge 0.95 Sn 0.05 fins fabrication process are provided in the Methods section. Figure 7a shows the top-view scanning electron microscope (SEM) image of the as-patterned Ge 0.95 Sn 0.05 fins after dry etch. The fin width is ~29.8 nm. Digital etch (RF: 250 W, Oxidation time: 120 s) was performed on the sample, and the fin width obviously shrinks with increased etch cycle (Fig. 7b and c). The etch rate of ~2.4 nm/ cycle can be extracted from the SEM image, which is higher than that extracted from bulk Ge 0.95 Sn 0.05 sample. The different etch rate should arise from different crystal orientation of the etch surface. The patterned Ge 0.95 Sn 0.05 fins are along <110> direction, thus the oxidation and etch occur on the {110} oriented surfaces. Ligenza et al. suggested that the orientation effect on oxidation rate might be caused by differences in the dangling bond density on the various crystal surfaces 52 . As the {100} surfaces have lower dangling bond density than {110} surfaces, the oxidation rate or etch rate for {100} surfaces is slower. Figure 7d shows the tilted-view SEM image of the Ge 0.95 Sn 0.05 fins after 5-cycle digital etch. Extremely scaled Ge 0.95 Sn 0.05 fins were produced with good uniformity. The region where the transmission electron microscope (TEM) lamella was prepared by focused ion beam (FIB) milling is highlighted by a green solid line along CC' . The cross-sectional TEM image of the Ge 0.95 Sn 0.05 fins is shown in Fig. 7e. Trapezoidal fins with narrow upper base were formed. The high resolution TEM (HRTEM) image of one Ge 0.95 Sn 0.05 fin is presented in Fig. 7f, demonstrating the high crystalline quality of the Ge 0.95 Sn 0.05 fins. The fin width was trimmed down to 5 nm after 5-cycle digital etch. The side wall of Ge 0.95 Sn 0.05 fin is smooth with no observed crystal damage as shown in the inset of Fig. 7f.
In summary, we have demonstrated a new digital etch technique for controllable etching of Ge or Ge 1−x Sn x materials with high precision. The two-step digital etch approach consists of Ge 1−x Sn x oxide formation by low temperature plasma oxidation and the oxide removal in diluted HCl. The plasma oxidation is a good self-limiting process, as the oxide layer thickness grows logarithmically with the oxidation time and saturates fast. Consistent etch rate can be obtained on the Ge 1−x Sn x samples, independent of the initial Sn content. After several cycles of digital etch, the surface morphologies and RMS roughness do not show discernible changes as compared to the as-grown samples. Extremely scaled Ge 0.95 Sn 0.05 fins with 5 nm fin width were demonstrated with the digital etch technique. The Ge 0.95 Sn 0.05 fins show high crystalline quality, with smooth side wall. The unique process capability can facilitate fabrication of Ge or Ge 1−x Sn x based nanostructures with high-precision control for a range of CMOS, Si-compatible photonics and microelectro-mechanical (MEMS) device applications.

X-ray photoelectron spectroscopy (XPS).
Measurements were performed using a VG ESCALAB 220i-XL imaging XPS system. Monochromatic aluminum (Al) Kα X-ray (1486.7 eV) was employed with the photoelectrons collected at a take-off angle of 90° (with respect to the sample surface). After Shirley-type background subtraction, the normalized peak area is calculated by taking into consideration of the corresponding Scofield photoionization cross-sections (SF), the transmission function of the spectrometer by the manufacturer (TXFN) and the energy compensation factor. The atomic concentration of Sn, x, is calculated using

Sn Sn Ge
where A Sn is the normalized Sn peak area and A Ge is the normalized Ge peak area.
High-resolution X-ray diffractometry (HRXRD). HRXRD was used to investigate the crystalline quality, composition, and strain of the samples by using the X-ray demonstration and development (XDD) beamline at the Singapore Synchrotron Light Source (SSLS). The wavelength of incident X-ray beam was 0.1634 nm. The diffractometer is the Huber 4-circle system 90000-0216/0, with high-precision 0.0001° step size for omega and two-theta scans. The storage ring, Helios 2, was running at 700 MeV with typical stored electron beam current of 300 mA. X-ray reflectivity measurements were also done at the SSLS.
Tapping mode atomic force microscope (AFM, Bruker Dimension FastScan). AFM was employed to characterize the surface morphology and roughness of the samples before and after digital etch. The background slope was removed by subtracting a first order plane fit, and the tilt was removed by a first order flatten with the NanoScope Analysis software.