Knowledge-Based Evolutionary Linkage in MEMS Design Synthesis

TLDR: This work focuses on CBR, a knowledge-based algorithm, and MOGA to examine how biological analogs that exist between the authors' evolutionary system and nature can be leveraged to produce new promising MEMS designs.

Abstract: Multi-objective Genetic Algorithms (MOGA) and Case-based Reasoning (CBR) have proven successful in the design of MEMS (Micro-electro-mechanical Systems) suspension systems. This work focuses on CBR, a knowledge-based algorithm, and MOGA to examine how biological analogs that exist between our evolutionary system and nature can be leveraged to produce new promising MEMS designs. Object-oriented data structures of primitive and complex genetic algorithm (GA) elements, using a component-based genotype representation, have been developed to restrict genetic operations to produce feasible design combinations as required by physical limitations or practical constraints. Through the utilization of this data structure, virtual linkage between genes and chromosomes are coded into the properties of pre-defined GA objects. The design challenge involves selecting the right primitive elements, associated data structures, and linkage information that promise to produce the best gene pool for new functional requirements. Our MEMS synthesis framework, with the integration of MOGA and CBR algorithms, deals with the linkage problem by integrating a component-based genotype representation with a CBR automated knowledge-base inspired by biomimetic ontology. Biomimetics is proposed as a means to examine and classify functional requirements so that case-based reasoning algorithms can be used to map design requirements to promising initial conceptual designs and appropriate GA primitives. CBR provides MOGA with good linkage information through past MEMS design cases while MOGA inherits that linkage information through our component-bsased genotype representation. A MEMS resonator test case is used to demonstrate this methodology.

Figures

Fig. 3. MEMS resonator gene representation [6]

Fig. 7. Resonant frequency = 23.8 kHz for initial MOGA design (a); Resonant frequency = 24.8 kHz for a pareto optimal y-symmetric design generated by MOGA (b)

Fig. 11. Best performing designs based on frequency deviation for D1, D2, D3, and D5

Fig. 6. MEMS design synthesis architecture

Fig. 9. Examples of MEMS resonator designs with increasing constraints

Table 3. Comparison of design area and frequency deviation

Full text

UC Berkeley
Green Manufacturing and Sustainable
Manufacturing Partnership
Title
Knowledge-Based Evolutionary Linkage in MEMS Design Synthesis
Permalink
https://escholarship.org/uc/item/3nt1r0zh
Authors
Cobb, Corie L.
Zhang, Ying
Agogino, Alice M.
et al.
Publication Date
2008
eScholarship.org Powered by the California Digital Library
University of California

Y.-p. Chen, M.-H. Lim (Eds.): Linkage in Evolutionary Computation, SCI 157, pp. 461–483, 2008.
springerlink.com
© Springer-Verlag Berlin Heidelberg 2008
Knowledge-Based Evolutionary Linkage in MEMS
Design Synthesis
Corie L. Cobb
1
, Ying Zhang
2
, Alice M. Agogino
1
, and Jennifer Mangold
1
1
Mechanical Engineering Department at the University of California,
Berkeley, CA 94720, USA
ccobb@berkeley.edu, agogino@berkeley.edu, jam@me.berkeley.edu
2
School of Electrical and Computer Engineering at the Georgia Institute of Technology,
Savannah, GA 31407, USA
yzhang@gatech.edu
Abstract. Multi-objective Genetic Algorithms (MOGA) and Case-based Reasoning (CBR)
have proven successful in the design of MEMS (Micro-electro-mechanical Systems) suspen-
sion systems. This work focuses on CBR, a knowledge-based algorithm, and MOGA to exam-
ine how biological analogs that exist between our evolutionary system and nature can be lever-
aged to produce new promising MEMS designs. Object-oriented data structures of primitive
and complex genetic algorithm (GA) elements, using a component-based genotype representa-
tion, have been developed to restrict genetic operations to produce feasible design combinations
as required by physical limitations or practical constraints. Through the utilization of this data
structure, virtual linkage between genes and chromosomes are coded into the properties of pre-
defined GA objects. The design challenge involves selecting the right primitive elements, asso-
ciated data structures, and linkage information that promise to produce the best gene pool for
new functional requirements. Our MEMS synthesis framework, with the integration of MOGA
and CBR algorithms, deals with the linkage problem by integrating a component-based geno-
type representation with a CBR automated knowledge-base inspired by biomimetic ontology.
Biomimetics is proposed as a means to examine and classify functional requirements so that
case-based reasoning algorithms can be used to map design requirements to promising initial
conceptual designs and appropriate GA primitives. CBR provides MOGA with good linkage in-
formation through past MEMS design cases while MOGA inherits that linkage information
through our component-bsased genotype representation. A MEMS resonator test case is used to
demonstrate this methodology.
1 Introduction
Microelectromechanical Systems (MEMS) are small micro-machines or micron-scale
electro-mechanical devices that are fabricated with processes adapted from Integrated
Circuits (ICs). Although still a relatively new research field, MEMS devices are being
developed and deployed in a broad range of application areas, including consumer
electronics, biotechnology, automotive systems and aerospace. Example MEMS
devices include accelerometers in automotive airbags and micro-mirrors for optical
switching in data communication networks. As MEMS devices grow in complexity,
there is a greater need to reduce the amount of time MEMS designers spend in the

462 C.L. Cobb et al.
initial conceptual stages of design by employing efficient computer-aided design
(CAD) tools.
Working with a multidisciplinary research team at the Berkeley Sensor and Actua-
tor Center (BSAC), our work with Evolutionary Computation (EC) is focused on the
conceptual design of MEMS devices. Zhou et al. [1] were the first to demonstrate that
a multi-objective genetic algorithm (MOGA) can synthesize MEMS resonators and
produce new design structures. SUGAR [2], a MEMS simulation tool, was used to
perform function evaluations on constraints and fitness values. Kamalian et al. [3] ex-
tended Zhou’s work and explored interactive evolutionary computation to integrate
human design expertise into the synthesis process. They also fabricated and tested the
emergent designs in order to characterize their mechanical properties and identify
deviations between simulated and fabricated features [4]. Zhang et al. [5, 6] imple-
mented a hierarchical MEMS synthesis and optimization architecture, using a compo-
nent-based genotype representation and two levels of optimization: global genetic
algorithms (GA) and local gradient-based refinement. Cobb et al. [7] created a case-
based reasoning (CBR) tool to serve as an automated knowledge base for the synthe-
sis of MEMS resonant structures, integrating CBR with MOGA [8] to select
promising initial designs for MOGA and to increase the number of optimal design
concepts presented to MEMS designers.
In related research, Muhkerjee et al. [9] conducted work on MEMS synthesis for
accelerometers using parametric optimization of a pre-defined MEMS topology. They
expanded the design exploration within a multidimensional grid in order to find the
global optimal solution. Wang's [10] approach to MEMS synthesis utilized bond
graphs and genetic programming with a tree-like structure of building blocks to in-
corporate knowledge into the evolutionary process, similar to work by Zhang [6]. Li
et al. [11] concentrated on developing automated fabrication process planning for sur-
face micromachined MEMS devices that relieves designers from the tedious work of
process planning so they can concentrate on the design itself. MEMS CAD has
matured to the point that there are now commercial CAD programs, such as Comsol®
and IntelliSuite®, that offer MEMS designers pre-made modules and cell libraries,
but there is little automatic reasoning in place for the user on how and when these
components should be used.
Our EC method employs a genetic algorithm as the evolutionary search and opti-
mization method. GAs were introduced by Holland [12] to explain the adaptive proc-
esses of evolving natural systems and for creating new artificial systems in a similar
way, and Goldberg [13] further demonstrated how to use them in search, optimiza-
tion, and machine learning. Chen et al. [14] noted that traditional GAs require users to
possess prior domain knowledge in order for genes on chromosomes to be correctly
arranged with respect to the chosen operators. The performance of a GA is heavily
dependent upon its encoding scheme. When prior domain knowledge is available, the
design problem can be solved using traditional genetic algorithms. However, that is
not always the case, and this is when methods such as linkage learning are needed.
Chen [15] and Harik [16] both focused research efforts on the linkage learning genetic
algorithm (LLGA) so that a GA, on its own, can detect associations among genes to
form building blocks [15].

Knowledge-Based Evolutionary Linkage in MEMS Design Synthesis 463
Linkage is an important part of GA performance. Tightly linked genes are syn-
onymous with building blocks, but higher level linkage amongst building blocks is
also necessary to ensure successful design solutions are reached. We propose an inte-
grated MEMS design synthesis system which combines CBR with biologically
inspired classifications and an evolutionary algorithm, MOGA, to help generate more
varied conceptual MEMS design cases for a designer and her/his current design
application.
In this chapter, we will explain our micro-resonator test case which will be high-
lighted throughout our work to explain our linkage concept. Next, we discuss MOGA
and CBR and explain how linkage is achieved through our knowledge-based evolu-
tionary algorithm. Lastly, we present a review of symmetry patterns observed in
nature, as they pertain to resonant frequency-sensitive biological creatures, and
explore the role that symmetry plays in our evolutionary synthesis process for the
resonator example.
2 Evolutionary Computation for Resonant MEMS Design
2.1 MEMS Resonator Test Case
To date, our MEMS design synthesis program has focused on the design of resonant
MEMS. A schematic of a MEMS resonator and its component decomposition are
shown in Fig. 1. These designs have consisted of a fixed center mass (either with or
without electrostatic comb drives) connected to four ‘legs’, each made up of multiple
beam segments. We evaluated our MOGA synthesis program for several sets of
performance objectives all calculated using the SUGAR simulation program.
Fig. 1. Schematic of example resonator synthesis problem. The geometry of the center mass is
fixed, while the number of beam segments per leg and the size and angle of each segment is
variable [3].
As we are designing resonators, the most significant performance objective for all
structures is the resonant frequency (f
0
). Resonant frequency is the most critical
requirement because if a resonator deviates too far from its frequency target it is es-
sentially a useless design. Other performance objectives we have used for synthesis

464 C.L. Cobb et al.
include the stiffness of the structure in the x or y-direction as well as the device area
(defined by a bounding rectangle around the device).
2.2 SUGAR: MEMS Simulation with Modified Nodal Analysis
SUGAR [2] is an open-source MEMS simulation tool based on modified nodal analy-
sis (MNA), allowing a designer to quickly prototype and simulate several complex
MEMS structures for preliminary design applications.
1
Finite element analysis (FEA)
calculations could take hours per simulation, making them infeasible for iterative
design processes on complex systems. SUGAR and other similar lumped parameter
nodal analysis simulation tools can perform these functional calculations with reason-
able accuracy at a fraction of the time and can therefore allow the MEMS designer to
explore larger design spaces. FEA and parametric optimization can then be used
to refine the most promising of the design concepts produced by the MOGA evolu-
tionary process.
2.3 Linkage with Component-Based Genotype Representation
Genetic linkage, in biological terms, refers to the relative position of two genes on a
chromosome. Two genes are linked if they are on the same chromosome and are
tightly linked if they are physically close to each other on the same chromosome.
Genes that are closely linked are usually inherited together from parent to offspring
[14]. Our MOGA data structure can be classified as “linkage adaptation” if we use the
same terminology as Chen [14]. Linkage adaptation refers to specifically designed
representations, operators, and mechanisms for adapting genetic linkage along with
the evolutionary process. Chen states that linkage adaptation techniques are closer to
biological metaphors of evolutionary computation because of their representations,
operators, and mechanisms.
Fig. 2. Gene representation examples for MEMS building blocks [6]
1
SUGAR can be accessed from:
http://sourceforge.net/projects/mems/

Frequently asked questions

Q1. What can be used to refine the promising of the design concepts produced by the MOGA evolutionary?

FEA and parametric optimization can then be used to refine the most promising of the design concepts produced by the MOGA evolutionary process. 

Q2. What are the contributions in "Title" ?

This work focuses on CBR, a knowledge-based algorithm, and MOGA to examine how biological analogs that exist between their evolutionary system and nature can be leveraged to produce new promising MEMS designs. Biomimetics is proposed as a means to examine and classify functional requirements so that case-based reasoning algorithms can be used to map design requirements to promising initial conceptual designs and appropriate GA primitives. 

Q3. What future works have the authors mentioned in the paper "Title" ?

As part of their future research plan, the authors will examine how linkage learning can be integrated with MOGA when CBR may not be able to select a good initial seed design. Further exploring biomimetic algorithms and biomimetic ties to MEMS synthesis algorithms is another area the authors plan to pursue, investigating how increasing the number of leg components on a MEMS design can create optimal solutions in other design areas such as micro-robots. The authors want to also further explore the role symmetry and angle constraints have on these types of new MEMS designs. Lastly, the authors are moving towards creating a broader MEMS classification scheme and building up a case library of MEMS filter designs and their accompanying components to further expand the range of designs covered by their program. 

Q4. What is the important performance objective for a resonator?

Resonant frequency is the most critical requirement because if a resonator deviates too far from its frequency target it is essentially a useless design. 

Q5. How many runs of the MOGA process were conducted for each constraint case?

Using constraint cases of (1) no symmetry, (2) y-axis symmetry, and (3) x- and y-axis symmetry, five runs of the MOGA process were conducted for each constraint case in order to see a good spread of design solutions. 

Q6. How many legs do you need to move in a resonator?

In their MEMS resonator design, the authors only want to move in one direction based on the comb drive actuation, hence four legs provides more balance and stability than two legs. 

Q7. What was the first successful application of CBR?

The first successful industry application of CBR was CLAVIER [20] which was used by Lockheed Martin for determining successful loads of composite material parts for curing in an autoclave. 

Q8. What is a general hierarchy or structure of ontology?

A general hierarchy or structure of ontology is the following [26]: objects, classes of objects, attributes of objects, and relations between objects. 

Q9. What is the main reason for the MOGA synthesis?

Incorporating other powerful computational tools, such as CBR, with MOGA can help MOGA converge faster and more efficiently to optimal design concepts. 

Q10. Why is it important to index cases by both?

Because the user of their CBR program may be searching for designs based on input and output domains or application areas, it is important to index cases by both. 

Q11. Why did MOGA determine that suspensions could produce a better resonant frequency and?

because frequency and stiffness were also part of the optimization problem, MOGA determined that a design with the suspensions outside of the mass could produce a better resonant frequency and stiffness ratio. 

Q12. What is the importance of vibration signals in MEMS?

Vibration signals are also important, because many of the insect’s or spider’s prey produce vibrations through movement or feeding, which enables them to be located more easily [33]. 

Q13. What is the role of symmetry constraints in the MEMS synthesis architecture?

Increasing the level of symmetry constraints can further restrict the search space to a more manageable sizeand enable their micro-resonator designs to achieve a smaller design area on average, but more asymmetrical designs are favored by MOGA for reducing frequency error and achieving the smallest design area. 

Q14. What are the two common categories of MEMS?

Sensors and actuators are the two most broad and commonly agreed upon categories of MEMS which can be divided further into families and classes. 

Q15. What is the effect of the stiffness ratio on the design?

This bias in the stiffness ratio potentially forces the designs generated by MOGA to favor more asymmetrical layouts (C1 and C2) rather than fully symmetrical results (C3 and C4). 

Q16. What constraint is used to design a resonant structure?

C4 includes a manhattan angle constraint and represents the typical constraints a human MEMS designer will impose upon the design of a resonant structure. 

Q17. What is the design for asymmetrical or bilateral symmetry?

If the authors examine their results more closely, the authors must note that most of their design requirements favor asymmetrical or bilateral symmetry if frequency is the major consideration and full symmetry if average area minimization over the pareto set is the priority. 

Q18. What is the difference between asymmetric and asymmetric animals?

In biology studies by Moller et al. [32], they found that growth rate and fluctuating asymmetry are negatively correlated, meaning asymmetric animals grow less rapidly than symmetric ones. 

Q19. How many legs can be desirable in a resonator?

if the authors look more broadly at other MEMS designs, such as micro-robots, more legs can be desirable to enable quick and easy movement in multiple directions. 

Q20. What is the definition of a MEMS device?

Although still a relatively new research field, MEMS devices are being developed and deployed in a broad range of application areas, including consumer electronics, biotechnology, automotive systems and aerospace. 

Q21. How many species of spiders have been classified?

Biologists have classified over 40,000 species of spiders, but they believe there are still thousands of species which have not yet been identified and named.