|
The goal of
design synthesis is to help engineers develop rapid, optimal configurations
for a given set of performance and constraint guidelines. Recent results in
the development of synthesis tools for MEMS will be presented as part of
the foundation for a larger research program aimed at developing a
practical synthesis tool that can create both the topology and sizing of
MEMS devices.
Multi-objective
Genetic Algorithms (MOGA) have been successfully implemented for MEMS
design in previous research [1]. This method was chosen because of its
generality, robustness and ability to optimize for multiple design goals.
MOGA uses an evolutionary approach to developing a population of optimal
solutions: Given a higher-level description of the device’s desired
behavior, an initial population of candidate designs is generated randomly from
a number of available components such as anchors, beams, electrostatic
gaps, combs and springs. Each design is checked for geometrical validity
and its performance is evaluated. MOGA is then applied to the initial
population to iteratively search for functional designs by applying the
genetic operations of selection, elitism, crossover and mutation to create
the next generation of designs. This process continues until an optimum
group of "pareto optimal" solutions is
synthesized. The modified nodal analysis (MNA) tool SUGAR is used to
evaluate the performance of the MEMS design. The use of SUGAR's
much quicker MNA method, rather then the relatively slow finite element
simulation, is critical to its computational tractability [2]. We will
summarize results using MOGA to synthesize MEMS devices and subassemblies
with a variety of combinations of performance objectives.
As part of our
evaluation of MEMS synthesis techniques, we will contrast MOGA against
Simulated Annealing (SA) optimization. As the name implies, SA exploits an
analogy between the way in which a heated metal cools into a minimum energy
state and a stochastic optimization algorithm that slowly “lowers the
temperature” in stages to eventually “freeze” at the global optimum. SA
randomly perturbs a given initial design, whose variations are accepted as
the new design with a threshold probability which decreases as the
computation proceeds. The slower the rate of probability decrease, the more
likely the algorithm is to find an optimal or near-optimal solution.
Simulated annealing will be compared to the genetic algorithm approach in
terms of robustness, effectiveness and speed. The research will be framed
within a larger research program for developing general-purpose MEMS
synthesis tools, including a case-based library for initial device designs.
References
1. N. Zhou, B.
Zhu, A.M. Agogino, K.S.J. Pister,
“Evolutionary Synthesis of MEMS (Microelectronic Mechanical Systems)
Design”. Proceedings of ANNIE 2001, Intelligent Engineering Systems through
Artificial Neural Networks, Volume 11, ASME Press, pp. 197-202.
2. J.V. Clark,
D. Bindel, N. Zhou, J. Nie,
W. Kao, E. Zhu, A. Kuo, K.S.J. Pister, J. Demmel, S. Govindjee, Z. Bai, M. Gu, and A.M. Agogino,
“Addressing the Needs of Complex MEMS Design”, Proceedings of the 15th IEEE
MEMS Conference, Las Vegas, January 20-24, 2002, pp 204-209.
|