Reliability prediction of single crystal silicon MEMS using dynamic Raman spectroscopy

Abstract

The work investigates an extension and improvement to reliability prediction in single crystal silicon MEMS by utilising dynamic Raman spectroscopy to allow fracture test data collected directly from devices thereby taking account of actual geometrical tolerances, dynamic load conditions and effects from the microfabrication process. Micro-cantilever beam MEMS devices microfabricated from (100) crystalline silicon wafers having [110] beam direction were used in this experiment. A piezo actuator was used to vibrate the devices. The fracture data was taken by increasing the supply voltage to bring each device to rupture whilst a continuous beam HeNe laser was directed from the [100] direction to particular positions on the sample allowing for capture of the 𝐿𝑂𝑧 photons. The resulting Raman profile, broadened by the vibration of the device, was fit using a Voigt profile and compared to the no load condition. A calibration step was used to convert the Raman signal to volumetric μstrain. The reliability prediction methodology used in this work was developed under the Weibull distribution function that is based on the concept of weakest link theory describing distribution of flaws in brittle material. As each device was fabricated from semiconductor grade single crystalline silicon, it can be considered to have no mechanical defects with the flaws on the device only existing as surface flaws induced from the microfabrication process during manufacturing. Differently processed surfaces each have their own Weibull parameters. The failure prediction of a particular MEMS device is calculated from these parameters with the simulated structure responses due to applied load being predicted from the finite element package ANSYS. The failure prediction method shows a good agreement with the experimental results with accuracy of 10%. However, visual observations were necessary as a number of ruptured specimens started to fail from the bottom side of the clamped end and propagated through [111] direction so from the upper side the failure looks like it started from a distance from the clamped end. The experimental work was carried out utilising [100] Raman scattering. This limits the ability of the system to only capture the strain condition in one direction; the other strain directions were approximated using silicon orthotropic material properties and assuming that the load is a uniaxial load. This limitation forces the failure prediction distribution function to treat silicon as an isotropic material with the strength ii characteristic scale parameter in the Weibull distribution being the same value for all directions. This limitation together with extending the work towards implementing an “off-axis” Raman characterisation able to characterise all the strain directions is discussed for future work.