Abstract :
[en] In the context of microelectromechanical systems, due to their reduced size, the surface
phenomena, e.g. adhesion contact, play an important role in the reliability of the devices.
Indeed, the adhesive forces, e.g. capillary and van der Waal forces, can lead to the
stiction failure for which the two contacting surfaces can accidentally be stuck together
permanently. This is a common failure of MEMS. Because of the comparability between
the roughness of the contacting surfaces and the ranges of adhesive stress, the interaction
area can be much smaller than the apparent one. Since the contact zone is reduced and
becomes comparable with the characteristic length scale of the surface roughness, the
behaviors of micro structures subjected to adhesion suffer from a scatter, i.e. while some
devices are unstuck, the others with an identical design are stuck. The objective of this
work is to predict in a probabilistic way the adhesion behaviors of MEMS by accounting
for the randomness of the contacting surfaces.
The straightforward solution toward this end consists in constructing a numerical
model, e.g. using finite element method, and in performing a Monte-Carlo simulation
(MCS) directly on that model. Because the problem spans multiple scales, including the
nanometers range of adhesive stress and the micrometers length scale of MEMS, that
method demands a huge computational cost and becomes unpractical. In this work, a
stochastic model-based multiscale method is developed to fulfill the predefined objective
while remaining efficient in terms of computational cost. In this model, MCS is also per-
formed, however, in a scale-by-scale way. With this method, the model is executed with
acceptable computational cost. To verify the proposed model, a comparison in terms
of the numerical predictions obtained from two approaches, direct MCS and stochastic
model-based method, is performed. Furthermore, the model is applied to simulate the
stiction tests reported in the literature, and also on the experimental surfaces fabri-
cated by our partner at IMT-Bucharest lab 1 (without stiction test). By comparing the
numerical predictions with the experimental results, the model is then validated.
The model is used to broaden our knowledge in stiction phenomenon by considering
the effects of the following aspects on the adhesion energies: the roughness of surfaces,
the non-Gaussianity in the probability distribution of surface heights, and the humidity
of the environment conditions. Furthermore, the comparison between different sources
of uncertainty, e.g. due to the surfaces roughness and in the geometry dimensions of the
devices, is performed.
