Raman spectroscopy as a mapping tool for localized strain in microengineered structures

The behaviour of microengineered devices is routinely modelled using finite element analysis (FEA). Whilst this is becoming increasingly sophisticated, the values of the material data used in FEA packages have considerable associated uncertainty. In particular, there is no general technique to measure localized strain: the FEA is done using an average value taken from, for instance, a displacement measurement followed by calculations on bending stresses. The aim of this paper is to show how Raman spectroscopy can be used to assess localized strain, by first using the technique to calibrate doping profiles in silicon: doping itself being a major contributor to the strain in a device. Localized results are important because many applications require heavily doped silicon structures: in high concentrations the dopant will distort the silicon lattice considerably. In addition, dopants are known to aggregate at certain crystal defects such as dislocations. Thus the distribution of the dopant, and the associated strain, will depend on the original crystal quality and its processing history prior to dopant incorporation. Results have previously been reported on Raman spectroscopy of boron doped silicon [1], using a destructive etch technique to obtain a through-wafer doping profile. To our knowledge no results have been published on localized measurement of doping with the spatial resolution (,1 μm) proposed here, and using a nondestructive technique. There is an additional benefit from the work. Some microengineered devices, in particular those which require resonant motion, may have long term failure mechanisms which are caused by localized strain maxima. The ability to map the distribution of strain within a device, and then correlate this data with points of failure, should lead to more reliable design and manufacturing. The principles of this technique have been reported elsewhere, e.g. [2], as have details of the microline focus spectrometer (MiFS) used in our experiments [3]. Fig. 1 is a schematic of the instrument. Both the band shift and band half-width can be plotted as a function of position to give a direct two-dimensional picture. The established strain induced shift in orientated single crystal silicon and other materials points to the use of shifts in the doped silicon phonon frequency to determine relative strain distributions in the fabricated structures. As a result, strain peaks in the device, including overlays, can be observed. The instrumentation is capable of obtaining spectra with sub-micron resolution and can generate profiles and images representing intensity (species concentration), frequency (stress) and bandwidth (crystallinity). Applied mechanical stresses will result in phonon frequency changes, and structural alterations will change the phonon density of states population and thus the band shape. In the case of silicon these spectral responses to stress have been published [4], although the manifestation of all the theoretically predicted spectral changes has not been observed. A (1 0 0) n-type silicon wafer, of 10 U cm resistivity, was used as the starting material. After cleaning and etching to remove the top 3 μm of the wafer surface, the substrate was loaded into a diffusion furnace alongside a solid source of boron. Diffusion was carried out at 1100 8C for two hours. This is enough to produce a heavily doped p-type surface layer, with a concentration in excess of 3 3 10 ions cmy3 at a depth of 2 μm; confirmation is provided by this process being used to produce an etch stop layer in an ethylenediamine pyrocatechol (EDP)-based process. Substantial changes to the phonon spectrum of silicon in boron-diffused material have been ob-