25-10-2012, 02:45 PM
Case-based Reasoning for Evolutionary MEMS Design
Case-based Reasoning for Evolutionary MEMS Design.pdf (Size: 550.18 KB / Downloads: 32)
Abstract
A knowledge-based computer-aided design tool for microelectromechanical systems (MEMS)
design synthesis called CaSyn-MEMS (Case-based Synthesis of MEMS) has been developed.
MEMS-based technologies have the potential to revolutionize many consumer products and
create new market opportunities in areas such as biotechnology, aerospace, and data
communications. However, the commercialization of MEMS still faces many challenges due to
a lack of efficient computer-aided design tools that can assist designers during the early
conceptual phases of the design process. CaSyn-MEMS combines a case-based reasoning
(CBR) algorithm and a MEMS case library with parametric optimization and a multi-objective
genetic algorithm (MOGA) to synthesize new MEMS design topologies that meet or improve
upon a designer’s specifications.
CBR is an artificial intelligence methodology that uses past design solutions and adapts
them to solve current problems. Having the ability to draw upon past design knowledge is
advantageous to MEMS designers, allowing reuse and modification of previously successful
designs to accelerate the design process. To enable knowledge reuse, a hierarchical MEMS case
library has been created. A reasoning algorithm retrieves cases with solved problems similar to
the current design problem. Focusing on resonators as a case study, case retrieval demonstrated
an 82% success rate. Using the retrieved cases, approximate design solutions were proposed by
first adapting cases with parametric optimization, resulting in a 25% reduction in design area on
average while bringing designs within 2% of the frequency goal. In situations where parametric
optimization was not sufficient, a more radical design adaptation was performed through the use
of a multi-objective genetic algorithm (MOGA). CBR provided MOGA with good starting
points for optimization, allowing efficient convergence to higher quantities of Pareto optimal
design concepts while reducing design area by up to 43% and meeting frequency goals within
5%.
Introduction
This paper presents a design synthesis tool, called CaSyn-MEMS, for early stage
Microelectromechanical Systems (MEMS) design. CaSyn-MEMS, which stands for Case-based
Synthesis of MEMS, is a computer-aided design (CAD) tool that assists MEMS designers with
concept development by utilizing past design structures to synthesize new structures. CaSyn-
MEMS integrates case-based reasoning (CBR), a knowledge reasoning algorithm, with
parametric optimization and a multi-objective genetic algorithm (MOGA). Using previously
Page 2 of 24
successful MEMS designs indexed in a hierarchical case library, CaSyn-MEMS synthesizes and
optimizes new design structures that meet a designer’s current set of design requirements. This
paper will demonstrate how CBR can support design ideation during the initial conceptual
phases of the design process, while enabling large stochastic search methods, such as genetic
algorithms (GAs), to converge to new promising design solutions.
MOGA algorithms developed previously [1-3] have proven successful in the design of
resonant MEMS structures. Zhang et al. [1] noted that seeding MOGA with a good initial design
is essential to helping MOGA converge to better design solutions in a practical number of
evolutions. One shortfall is that they worked with the same initial design for all of their
synthesis processes, limiting the range and quality of solutions their MOGA algorithm could
generate. In addition, the burden of selecting a seed design was placed on the user of the
MOGA algorithm. What previous studies lacked was an efficient automated knowledge-base
which removes from the human designer the burden of seeding the synthesis algorithm with
good starting designs. The CBR tool developed in this paper benefits MEMS design by giving a
wider range of good starting design cases for optimization and adaptation processes. The CBR
starting cases increase the quantity and quality of optimal design solutions synthesized by
evolutionary algorithms and enable convergence to a wider range of optimal design solutions.
Introduction to Microelectromechanical Systems (MEMS)
MEMS are micro-scale electronic and mechanical components made with fabrication technology
adapted from the field of Integrated Circuits. MEMS range from simple beams and electrostatic
gaps to more complex sensors and actuators that include fluidic, magnetic, and thermal systems.
MEMS can provide cost and size advantages as well as new functionality in products and
different application areas.
MEMS design efforts date back to the late 1960s when Nathanson et al. at Westinghouse
Research developed microscale resonant structures for filtering applications [4]. Today, micromechanical
elements along with circuitry can be combined together on a common silicon
substrate to create devices such as micro-accelerometers for deploying car air-bags [5], microactuators
for disk drives [6], micro-mirror arrays for projection display technology [7], and
microfluidic devices for ink-jet print heads [8].
Research Motivation
Maseeh et al. [9] surveyed MEMS companies and discovered the length of the product
development cycle was the most critical challenge to MEMS product commercialization. The
median time to develop a prototype device was 3.2 years with some taking as long as 8 years. In
particular, it is not uncommon for MEMS designers to rely on a “trial-and-error” method during
the initial stages of development. MEMS product development is an extremely costly and timeconsuming
process when compared with the average 14-27 month product development cycles
of consumer products [10].