Contactless monitoring of Ge content and B concentration in ultrathin single and double layer Si_1-xGe_x epitaxial films using multiwavelength micro-Raman spectroscopy Open Access

Stress and strain engineering is being actively pursued for further device performance enhancement.^1–3 During the last decade, the application of uniaxial strain in complimentary metal-semiconductor-oxide (CMOS) channels, by introducing Si_1-xGe_x alloys in the source/drain (S/D) region, has been implemented in advanced, high performance devices to overcome the limited scaling of gate length and oxide thickness.^3,4 For continuing improvement of strain engineering, reliable, non-destructive in-line material property monitoring and process diagnostic metrologies must be developed.

Channel stress is modulated strongly by the device structure including thickness, Ge content and dopant (typically B) concentration of the Si_1-xGe_x layer in the S/D region. Ge content, B concentration, and thickness of the Si_1-xGe_x have been measured by high resolution X-ray diffraction (HRXRD) and secondary ion mass spectroscopy (SIMS). Due to the destructive and time consuming nature of the SIMS process, a very limited number of samples were characterized.^3,5–7 As a nondestructive Ge content and Si_1-xGe_x film thickness monitoring technique, HRXRD has been utilized for the characterization of Si_1-xGe_x epitaxal thin film in Ge content.⁸ However, very poor spatial resolution, due to a very large blanket area requirement (>100 x 100 μm²) and a very high B detection limit of 2.7 x 10¹⁹ atoms/cm³, limit the usefulness of this technique. It is particularly not as effective for advanced device research and manufacturing, in which multi-layer Si_1-xGe_x epitaxial layers are in smaller patterns and/or B concentration, below the (HRXRD) detection limit, is frequently used. Typically, measurement was done at the center of the 300 mm diameter wafer and so, within wafer uniformity information could not be reliably collected.

As device scaling shrinks overall dimensions, Ge content and B concentration have been increasing, while thickness of Si_1-xGe_x epitaxial film stacks are getting thinner. Development of nondestructive, in-line monitoring of Ge content and B concentration of stacked Si_1-xGe_x epitaxial layers is highly desired.

Micro-Raman spectroscopy has long been used as a nondestructive, local, Si stress characterization technique for device structures.^9–17 It also enables high spatial resolution and depth profiling capabilities. It is well known that the probing depth of Raman measurements can be adjusted by selecting appropriate excitation wavelengths, because the absorption coefficient of a probing laser beam is a function of wavelength (or photon energy). Multiwavelength, high spectral resolution, micro-Raman spectroscopy^{5–7,14–17} was chosen as a most promising candidate for nondestructive, in-line diagnostic metrology for Ge content and B concentration analyses of single and double Si_1-xGe_x epitaxial layers. In addition, Raman spectroscopy was made significantly more stable and repeatable, for these applications through the development of a novel, polychromator-based, long focal length system.¹⁵ Many recent publications on Raman characterization of epitaxial Si_1-xGe_x on Si and channel stress characterization of device structures reflect a growing interest in this metrology.^3,5,7,11,16 Prior publications focus on characterization of single layer Si_1-xGe_x epitaxial layers on blanket Si wafers and large blanket monitoring pads (up to 300 x 300 μm²) on device wafers. In this paper, small size pads (50 x 80 μm² or smaller) of single and double layer Si_1-xGe_x epitaxial layers on Si device wafers were characterized by multiwavelength micro-Raman spectroscopy and compared with HRXRD and SIMS characterization results.

Undoped and B-doped Si_1-xGe_x epitaxial layers, grown on Si(100) device wafers, were prepared. A variety of Ge content, B concentrations, thicknesses and structures of Si_1-xGe_x epitaxial layers were grown on Si(100) device wafers and characterized using the polychromator-based, long focal length (2.0 m) Raman spectroscopy system shown in Fig. 1. The Ge content was varied from 15 ∼ 45 at%. The target B-doping level was in the range of 1.0 x 10¹⁸ ∼ 1.0 x 10²¹ atoms/cm³. The thicknesses of single and double Si_1-xGe_x epitaxial layers were varied in the range of 5 ∼ 120 nm. For the double Si_1-xGe_x epitaxial layers, the Ge content, B doping level and thickness of each layer was varied. Five different excitation wavelengths (363.8, 441.6, 457.9, 488.0 and 514.5 nm), from ultra-violet (UV) to visible, were used. The spectral resolution of the as-measured Raman spectra (before curve fitting) was in the range of 0.25 (at 363.8 nm) ∼ 0.07 cm^-1 (at 514.5 nm).

For probing different depths of Si_1-xGe_x multilayers, an appropriate excitation wavelength must be selected. Two Si_1-xGe_x monitor pads (50 x 80 μm²) on the patterned wafer were measured using micro-Raman spectroscopy under multiwavelength excitation. Figures 2(a)-2(c)show probing depths of the Raman measurements under the five different excitation wavelengths, an optical microscopy image of the monitoring pads and a cross sectional illustration of monitoring pads.

A HRXRD system (Jordan Valley Semiconductors’ JVX 7200) was used in this study. The nominal spot size of X-ray beam is ∼100 x 100 μm², which is significantly larger than the monitoring pad (50 x 80 μm²). The HRXRD data was modeled using Jordan Valley Semiconductors’ single layer and double layer analysis program to estimate the Ge content and Si_1-xGe_x epitaxial layer thickness.

Figure 3 shows Raman spectra of a 50 nm thick Si_1-xGe_x epitaxial layer on Si(100) wafers. Undoped and B-doped Si_1-xGe_x epitaxial layers with Ge content of target 20 at% and 32 at% were measured under the five excitation wavelengths selected. The target B concentration of doped Si_1-xGe_x epitaxial layers was ∼1.0 x 10¹⁹ atoms/cm³. For simplicity, Raman spectra measured under 363.8, 457.9 and 514.5 nm excitation were chosen for the figure. UV Raman spectra, under 363.8 nm excitation only, showed a single peak, corresponding to the Si-Si peak from the Si_1-xGe_x epitaxial layer, due to the shallow probing depth (∼5 nm). The undoped and B-doped Si_1-xGe_x epitaxial layer, with a target Ge content of 20 at%, showed a Si-Si peak at approximately ∼513 cm^-1 and ∼511 cm^-1, respectively. Raman spectra measured under visible excitation wavelengths (457.9 and 514.5 nm) with deeper probing depths showed two peaks (a weak Si-Si peak near ∼513 cm^-1 and a strong Si peak around ∼520.3 cm^-1). The longer excitation wavelengths result in a weaker Si-Si peak due to smaller contributions from the thin Si_1-xGe_x epitaxial layer. Since the downward shift of the Si-Si peak from the Si peak of 520.3 cm^-1 is linearly proportional to the Ge content (-0.35 cm^-1/ Ge at% in the case of fully strained, undoped Si_1-xGe_x),^5–7 the Si-Si peak position of ∼513 cm^-1 corresponds to Ge content of 20 at%. It is consistent with the targeted Ge content and agrees with the Ge content determined by SIMS. In the case of heavily B-doped Si_1-xGe_x, with lower Ge content, the Si-Si Raman shift is slightly smaller due to significant strain relaxation. For both undoped and B-doped cases, Si_1-xGe_x epitaxial layers, with a targeted Ge content of ∼32 at%, showed a Si-Si peak at ∼ 507.0 cm^-1. The B-doped Si_1-xGe_x epitaxial layer showed significant weakening and full-width-at-half-maximum (FWHM) broadening of the Si-Si peak, regardless of Ge content and excitation wavelength.

Figure 4(a) shows the correlation of Ge content measured by Si-Si Raman peak positions (under 457.9 nm excitation), as well as HRXRD and SIMS from the small Si_1-xGe_x layer pads of device wafers, with a nominal Ge content of 32 at%, without B doping. The flow rate of the GeH₄ precursor was varied by ±10% from standard flow rate, to observe the sensitivity of Raman measurements for in-line Ge content monitoring applications. The Si-Si Raman peak position is very sensitive and correlates perfectly to the Ge content determined by SIMS with a very narrow range of 31.0 ∼ 33.0 at %, while the Ge content determined by HRXRD showed very poor correlation. Ge content estimated by HRXRD showed very poor linearity as well as a significant offset (3∼5 at%) from SIMS depth profiling results. It has significant limitations as an effective in-line Ge content and B concentration monitoring technique for advanced devices which require tighter Ge and B uniformity and repeatability control.

B-doped Si_1-xGe_x layers, with a wide range of B concentrations, were grown by varying the flow rate of the GeH₄ precursor. FWHM values of the Si-Si Raman shift of B-doped Si_1-xGe_x layer pads are plotted in Fig. 4(b) as a function of B concentration, as determined by SIMS. FWHM widens linearly as B concentration in Si_1-xGe_x is increased, over a wide range of B concentrations (∼1.0 x 10¹⁸ ∼ 5.0 x 10²⁰ atoms/cm³). The detection limit of Raman measurement of B concentration in the Si_1-xGe_x layer was ∼7.8 x 10¹⁷ atoms/cm³, which is >30 times more sensitive than HRXRD has ever reported.⁸ 

In general, shift and FWHM of Si-Si Raman peak can be affected by many factors such as crystallinity, strain, stress and dopants in Si_1-xGe_x/Si. Because of these reasons, it is difficult to determine Ge content solely from the Si-Si peak shift and B concentration only from the FWHM of the Si-Si Raman peak. However, the ultra thin epitaxial Si_1-xGe_x/Si, with Ge content and thickness up to ∼35at% and ∼60 nm used in advanced devices, can be treated as fully strained films. Within these boundary conditions, the shift and FWHM of the Si-Si Raman peak from Si_1-xGe_x/Si can be used to estimate Ge content and B concentration in the Si_1-xGe_x layer.

Reliable Ge content measurement of small monitoring pads (smaller than 300 x 300 μm² in area) using HRXRD is very difficult, owing to the large beam size employed (∼100 x 100 μm²). The Ge content of small monitoring pads estimated by HRXRD, can easily be off by several at% from typical SIMS results, and so, it is difficult to use, with confidence, as a reliable in-line process/material monitoring technique. The poor sensitivity and constraints in measuring monitoring pads of different areas (ie. only those significantly larger than 100 x 100 μm²) required to use HRXRD to measure B doping, also makes it less effective for monitoring the Si_1-xGe_x multi-layers with large Ge content and B concentration variations, which are used in advanced devices at the 40 nm node and beyond. In contrast, the Ge content and B concentration of Si_1-xGe_x epitaxial layers, as thin as 5 nm, can be quantitatively and nondestructively characterized with high spatial resolution (probing area of <1.0 μm diameter) by this architecture for Raman spectroscopy.

The multiwavelength capability of this approach has powerful implications. It allows Raman characterization to be extended to stress/strain engineered double Si_1-xGe_x layer pads. The thicknesses of the double Si_1-xGe_x layers were 75 nm (top layer: 20 nm, bottom layer: 55 nm). The Ge content and B dopant concentration of top and bottom Si_1-xGe_x epitaxial layers were varied. The nondestructive characterization of the bottom Si_1-xGe_x epitaxial layer is particularly challenging. Normalized, multiwavelength Raman spectra from three wafers with double Si_1-xGe_x epitaxial layers are plotted in Fig. 5. The Ge content and B concentration in the bottom Si_1-xGe_x epitaxial layer were modified among the three wafers. The UV (363.8 nm) Raman spectra from all three wafers are almost identical. The positions of the very broad Si-Si peaks are at ∼512 cm^-1. This suggests the following: i) the thickness of the top layer of Si_1-xGe_x epitaxial layer is thicker than the probing depth (∼5 nm) of UV excitation, ii) the Ge content of the top Si_1-xGe_x layer is 15.5 at%, and iii) the B concentration is approximately 1.2 x 10²⁰ cm^-3. The largest variations in the Si-Si peaks were observed under 441.6 nm, suggesting significant variations in the Ge content and B concentration in the bottom Si_1-xGe_x epitaxial layers, among the three wafers. The Raman spectra from both 441.6 and 457.9nm excitation indicate the Ge content and B concentration ranges of 30.8 ∼ 31.1 at% and 4.1 ∼ 4.8 x 10¹⁹ cm^-3 for the bottom Si_1-xGe_x epitaxial layers for the three wafers. The B₂H₆ flow rate increase results in increase in both B concentration and Ge content in the Si_1-xGe_x epitaxial layer. It is due to the change of mole fractions of Si, Ge and B sources to the total gas flow and their interactions during epitaxial growth. As the excitation wavelength becomes longer (the probing depths becomes deeper), the Si-Si peak becomes significantly weaker and less sensitive to the variations in the bottom Si_1-xGe_x epitaxial layers. The Ge content and B concentrations of the double layer B-doped Si_1-xGe_x on Si, estimated by 363.8, 441.6 and 457.9 nm Raman spectra, showed good agreement with SIMS depth profiling results (Fig. 6). Raman spectra under 441.6 and 457.9nm excitation are more sensitive to Ge content and B concentration in the bottom Si_1-xGe_x layer than SIMS despite its non-contact and non-destructive nature of characterization.

For comparisons, HRXRD rocking curves measured from the same set of wafers are summarized in Fig. 7. All curves look almost identical and content be distinguished. Several small satellite peaks originated from Si_1-xGe_x layers with 5∼20 counts were measured in the left side of a very strong Si substrate peak with ∼10,000 counts. HRXRD cannot distinguish the difference between wafers with different Ge content and B concentration in the bottom layer. Even though HRXRD can give rough estimates of Ge content and B content above 2.7x10²⁰ cm^-3 in a single Si_1-xGe_x layer, it has severe limitation in monitoring Ge content and B concentration in multilayer Si_1-xGe_x/Si used in advanced device technology nodes beyond 40 nm and beyond.

The Ge content and B concentration in single and double Si_1-xGe_x epitaxial layers on Si(100) device wafers was investigated using high-resolution, multiwavelength micro-Raman spectroscopy to evaluate the feasibility of nondestructive characterization of Ge content and B concentration for in-line monitoring and epitaxial process optimization applications. The multiwavelength nature of this Raman spectroscopy architecture is especially relevant for the characterization of single or multi layer structures of this sort. HRXRD and SIMS characterization results are compared with multiwavelength Raman characterization results. HRXRD did not show sufficient sensitivity to detect B concentration of 1.2 x 10²⁰ cm^-3 or good correlation in Ge content with SIMS depth profiling results.

The Ge content and B concentration were estimated from the position and full-width-at-half-maximum of the Si-Si peak using multiwavelength Raman spectra from the single and double Si_1-xGe_x epitaxial layers. The values are in good agreement with SIMS depth profiling results. The feasibility of non-destructive, in-line monitoring of Ge content and B concentration of single and double Si_1-xGe_x epitaxial layers on small area monitoring pads was successfully demonstrated. Multiwavelength, high resolution Raman spectroscopy is very promising as a nondestructive, in-line monitoring technique for epitaxial process optimization, material property verification, and quality control applications.

The authors would like to thank Chih-Mou Huang, Ruey-Lian Hwang and Fu-Yuan Tseng of the TSMC FAND Department of HRXRD and SIMS analysis.