Nanoscale visualization of local strain

Strained silicon is promising as a substrate for metal-oxide-semiconductor transistors in future ultra-large-scale integrated circuits, because the carrier mobility is enhanced in this material compared to ordinary silicon substrates.¹ Accordingly, many methods of fabricating strained silicon on Si_1-xGe_x buffer layers have been reported.² To precisely control the properties of the transistors, investigation of the localized strain in such strained-silicon layers is required.

Raman spectroscopy has been widely used for the investigations of strain^3,4 because it nondestructively measures lattice vibrations (phonons) that are sensitive to strain. However, the low efficiency of Raman scattering makes it hard to observe localized strain at the nanometer scale. Moreover, for strained silicon grown on a Si_1-xGe_x substrate, Raman scattering from Si-Si vibrations in the thick (∼1μm) Si_1-xGe_x buffer layer can easily overwhelm the desired signal from Si-Si vibrations in the thin strained silicon layer (∼30nm) on top. Because of these two requirements for nanometer-scale sensitivity and surface selectivity, surface-enhanced Raman spectroscopy (SERS)⁵ and tip-enhanced Raman spectroscopy (TERS)^6–9 are exceptionally promising spectroscopic techniques.

Figure 1 compares conventional Raman spectroscopy on strained silicon with SERS, in which an 8nm-thick silver island layer acts as a surface enhancer. In the case of conventional Raman spectroscopy, the phonon mode from the strained silicon is almost overwhelmed by the strong background signal from the thick Si_1-xGe_x buffer layer. In contrast, SERS selectively enhances the phonon mode from the thin strained-silicon layer, resulting in a stronger signal than that from the buffer layer.¹⁰ However, the SERS technique is still destructive, because the substrate is coated with silver. Furthermore, the spatial resolution is strictly diffraction-limited.

Motivated by the SERS effect, we use TERS within a scanning system. The enhancement mechanism for TERS is similar to that for SERS, except that a nanometer-scale point source, namely a metallic tip, is used to provide high spatial resolution. Scanning the tip or sample then lets the technique be used as a microscope.

Figure 2 shows the concept of reflection-mode TERS microscopy for an opaque sample such as strained silicon.¹¹ Light, which is p-polarized in the same direction as the tip axis, illuminates the metallic tip and sample so that an enhanced electric field is generated at the tip apex due to surface-plasmon-polariton excitations. This enhanced electric field is used as a light source for nanometer-scale Raman spectroscopy. The spatial resolution is determined by the extent of the enhanced electric field, corresponding to the size of the tip apex.

Figure 3 compares TERS and conventional Raman spectra of strained silicon. As in the SERS results in Figure 1, TERS enhances the signal from the thin strained-silicon layer, but because it is surface selective it does not enhance the signal from the underlying Si_1-xGe_x buffer layer. Note that the enhanced volume in TERS (diameter∼30nm, corresponding to the tip size) is much smaller than the volume of SERS (diameter∼500nm, corresponding to the diffraction-limited focused spot). Nevertheless, TERS exhibits an enormous signal increase, comparable to that in SERS.¹²

To quantify the local stress, we use its linear relation to the deviation of the Raman-shift from that in normal silicon,

where f_ε-Si represents the stress, in Pa, and ω_Si (520cm^-1) and ω_ε-Si are the Si-Si phonon-mode frequencies, in cm^-1, of normal and strained silicon (_ε-Si) respectively.⁴ This equation means that 250MPa tensile stress shifts the spectral peak down by 1cm^-1 from its original position at 520cm^-1. The peak at 506.5cm^-1 observed in TERS corresponds to a local tensile stress of 3.375 GPa.

As mentioned above, one big advantage of TERS over SERS is the potential for scanning, which achieves nanometer-scale spatial resolution. Scanning the sample stage while detecting a TERS spectrum such as that in Figure 3 at each position visualizes the local stress distribution on the strained silicon surface, as shown in Figure 4. We can clearly see the ripple pattern of stress on the surface, which is probably induced by a point defect of the strained-silicon substrate.

One idea for improving the sensitivity of TERS for crystalline materials such as strained silicon is to use a depolarization detection scheme. In the case of crystalline material, the depolarized TERS signal can be separated from the strong background signal, which has different, well determined polarization.¹³ Another idea is to use UV illumination because, at 400nm wavelength, the penetration depth into silicon is only 5nm. However, we must also consider the optimal wavelength for the electric-field enhancement effect, which depends on the metal material, shape, and polarization of light.