Local strain characterization of MEMS-based silicon beams by Raman spectroscopy

State-of-the-art electronic devices are nowadays well into the nanometer regime. Raman spectroscopy is a non-destructive technique for chemical and mechanical analysis which can be used for local strain characterization of nanoscale devices. By measuring the inelastic scattering of light due to lattice vibrations and electronic excitations, the frequency shift in scattered light is compared with unstrained material to extract residual or applied strain. In this work, we show strain measurements in MEMS-based silicon beam structures fabricated following an original technique[1] which makes use of the internal stress present in an “as-deposited” silicon nitride actuator in order to induce controlled levels of deformation in the released silicon beams. The stress magnitude depends on the structure geometry and it enables the fabrication of structures arrays over a wide range of strain levels. The tensile strain was assessed by Raman spectroscopy using two lasers with a different penetration depth which allowed the study of strain both at surface and bulk levels. Excellent agreement of strain is shown across the entire range of beam lengths measured by Raman and compared with scanning electron microscopy (SEM). It is also demonstrated that important discrepancies may arise from the incorrect use of phonon deformation potentials (PDP) used to relate the Raman frequency shift with strain. Different PDPs have been reported for silicon[2-4]. In early experiments[2-3] the laser used was not able to penetrate through the entire material and the PDPs are therefore expected to be affected by relaxation of stress near the surface. To rectify this, later experiments[4] were carried out using a laser with greater penetration depth which enabled the extraction of more accurate PDPs. Nevertheless the migration from bulk silicon devices to nanostructures means that such PDPs are no longer appropriate. Furthermore, many groups currently use UV Raman with a penetration depth of roughly 10 nm to study thin layers. However, the PDPs chosen often remain those suitable for bulk silicon or are omitted from analysis. Our work highlights the need of an appropriate selection of PDPs for an accurate analysis of strain from Raman data. To date, no experimental values for PDPs suitable for strain determination in small devices have been reported. As a consequence some reports present results only in terms of Raman shift rather than strain while in other works a particular set of PDPs is arbitrarily chosen. Our work shows that in the case of silicon nanostructures, an improved agreement with theory is demonstrated by using older PDP values rather than the more recent values which relate primarily to bulk. Results indicate that in some cases errors up to 25% in strain determination are possible as a consequence of using a wrong set of PDPs.