Académique Documents
Professionnel Documents
Culture Documents
Author:
Romans Safonovs
Supervisors:
Jost Adam
ii
Preface
This work has been carried out as a part of Master of Science in Engineering - Mechat-
ronics programmme at the University of Southern Denmark, between September 2016
and June 2017.
I would like to thank my friends, parents and girlfriend for supporting me through-
out this process. Also, I would like to thank Fabian Lofink of Fraunhofer ISIT, for his
insight, experience and recommendations regarding fabrication of MEMS.
iii
Contents
List of Tables vi
1 Introduction 1
2 Theoretical Background 3
2.1 Magnetic Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Magnetic Field of a Coil . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Cantilever Beam Deflection . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Electric Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.6 MEMS Fabrication Processes . . . . . . . . . . . . . . . . . . . . . . . 9
6 Model Verification 26
6.1 Calculating the Magnetic Force . . . . . . . . . . . . . . . . . . . . . . 26
6.2 Calculating the Deflection . . . . . . . . . . . . . . . . . . . . . . . . . 28
iv
Contents
7 Modelling Efficiency 30
7.1 Exploiting the Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.2 Magnetic Force Calculation . . . . . . . . . . . . . . . . . . . . . . . . 31
7.3 Meshing Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Bibliography 45
B Additional Simulations 55
B.1 Cantilever Stress Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 55
B.2 Relation Between a Magnetic Film Length and a Magnetic Force . . . . 56
B.3 Optimal Coil Position Under a Magnetic Film . . . . . . . . . . . . . . 57
B.4 Coil Shape Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
C MATLAB Scripts 62
C.1 Magnetic Force Calculation . . . . . . . . . . . . . . . . . . . . . . . . 62
C.2 Result Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
v
List of Tables
vi
List of Figures
3.1 Switching principles for ohmic and capacitive devices. Source: [2] . . . 13
3.2 Concept scheme of a magnetic microactuator [19]. Source: [19] . . . . . 14
3.3 Concept scheme of a magnetic microactuator [20]. Source: [20] . . . . . 15
3.4 Concept scheme of a magnetic MEMS switch [22]. Source: [22] . . . . . 16
3.5 Concept scheme of a latching magnetic MEMS switch [23]. Source: [23] 17
3.6 Concept scheme of a latching magnetic MEMS switch [24]. Source: [24] 17
vii
List of Figures
6.1 Magnetic field in the center of a square electric loop due to 1 A of current 27
6.2 Force diagram of a cantilever beam under force Fmagnetic . . . . . . . . 28
viii
List of Figures
ix
1. Introduction
Every electric and electronic device takes use of switches: from flashlights and toasters
to mobile phones and power supplies. They allow or block current flow between two
conductors and are used to convert and manage energy and to route signals. And it
is logical that there is an ongoing research trying to find an ideal switch, which would
have zero on resistance, infinite off resistance, zero energy and time required to change
state, infinite lifetime, infinite power-handling capability and minimal size [1].
Of course, achieving such kind of characteristics is impossible due to laws of physics,
but it gives an idea, which parameters should be minimized and maximized and by
improving them, it would generally lead to a better power efficiency and smaller size
of any device that uses switches.
Efficiency is important, since it allows a system to accomplish the same task by
using less energy, which leads to many beneficial effects, such as lower operational
costs, lower pollution and others, depending on the device. For example, in case of
switching regulators, higher efficiency means less heat buildup, allowing for smaller and
lighter power supplies.
Previously, there were mainly two groups of switches: electromechanical relays
and solid-state relays. While electromechanical ones have better off-isolation, linearity
and low on resistance, solid-state switches are smaller, cheaper, faster and consume
less power [1]. But in recent years, micro-electro-mechanical systems (MEMS) have
become good candidates for combining good qualities of both above-mentioned switch
groups. There is a number of ways they can be operated: using electrostatic forces,
electrothermal expansion forces and magnetic forces. Electrostatically actuated MEMS
switches are easy to fabricate and hence, are the most common type of MEMS switches,
1
Introduction
but they are limited by their relatively high actuation voltage and small displacements.
Electrothermally actuated switches are the slowest type of switches, however they
produce the best contact forces , resulting in low contact resistance [1].
Magnetic actuation, however, is a promising approach to operate MEMS switches,
due to forces of higher magnitude and longer range than of electrostatic ones, even
though current designs are slower and consume a more power [2]. Operation mechanism
of such MEMS switches is illustrated below (Figure 1.1).
Figure 1.1: Operating principle of a magnetic MEMS switch proposed in this work
The goal of this work is to design and model a magnetic MEMS switch in such
a way that it is feasible in terms of fabrication capabilities of Fraunhofer Institute
of Silicon Technologies (Fraunhofer ISIT) and requires the least amount of power to
activate, possibly pushing the capabilities of such devices at microscale.
In the following chapters, first, I will cover concepts used throughout the project,
such as: magnetic force, magnetic field, cantilever beam deflection, electric power,
finite element method and microfabrication techniques. After that, I will explore state
of the art. Then, I will go though the process of modelling of a magnetic MEMS switch
(Figure 1.1) in COMSOL and compare the results to analytical solutions. And finally,
I will apply fabrication constraints of Fraunhofer ISIT and optimize all parameters of
the model for the smallest power dissipation. In the end, I will draw conclusions based
on the results and briefly sum up the work.
2
2. Theoretical Background
In this chapter I will briefly go through the most important concepts and formulas,
which were used or touched upon during this project, related to such fields as: magnetic
force, magnetic field, cantilever beam deflection, electric power, finite element method
and microfabrication techniques.
But Equation 2.1 only calculates the force acting only on a single loop. If there
is a magnetic material with volume V and magnetization per volume M , it can be
expressed as a collection of infinitely small loops, meaning that the total magnetic
force acting of a magnetic material in an external magnetic field is equal to [4]:
Z
Ftotal = ∇(M · B) dV. (2.2)
COMSOL calculates the force using the so-called Maxwell’s stress tensor S̄ [5]:
3
Theoretical Background
ZZ
1
Ftotal = dAS̄ · n (2.3)
µ0
where n is a unit vector pointing out of a unit surface dA, and S̄ is defined as:
Bx2 − B 2 /2 Bx By Bx Bz
1
S̄ = 2 2
By − B /2 (2.4)
By Bx By Bz
µ0
2 2
Bz Bx Bz By Bz − B /2
where B 2 = Bx2 + By2 + Bz2 .
µ0 Idl × ~x
dB = (2.5)
4π |~x|3
where dB, µ0 , I, dl and ~x are elemental flux density, current, vacuum permeability
(4π × 10−7 ), element of length and position vector respectively (Figure 2.1).
Using Equation 2.5 we can calculate magnetic fields due to a straight current wire
[4]:
4
Theoretical Background
µ0 I
B= (cos α1 + cos α2 ) (2.6)
4πd
where d, α1 and α2 are distance to the wire and angles that are shown on Figure 2.2
respectively. This equation can be expanded to calculate a magnetic field exerted by a
square loop simply by splitting the loop into 4 straight fragments, calculating magnetic
field for each one of them and adding the results together. See Appendix A.1 for a
detailed derivation of a magnetic field due to a rectangular electric loop.
Figure 2.2: Magnetic field B at point P due to an electric current I in a straight wire
COMSOL calculates the magnetic field by directly solving Gauss’s law for magnet-
ism and Ampere’s circuital law (static version, in this case):
∇·B =0 (2.7)
∇ × B = µ0 J (2.8)
In addition to calculating the magnetic force, it is also necessary to calculate how much
the cantilever bends under it in order to model a magnetic actuation of a switch in
Figure 1.1. The Euler-Bernoulli beam theory is a simplification of the linear theory of
elasticity, which calculates deflection of beams due to lateral loads [7]:
5
Theoretical Background
d2 d2 y
EI 2 = w(x) (2.9)
dx2 dx
where E, I, y, x and w are the Young’s modulus, second moment of inertia, deflection,
position and load respectively (Figure 2.3).
0=∇·S+F (2.10)
where S is stress tensor and F is external force. And finally, a strain tensor T is
calculated using the Hooke’s law for linear elastic materials [8]:
S = cT (2.11)
Power dissipation is an important figure of merit for any switch. It says how much
power is needed for it to actuate. Electric power P is defined as:
P =VI (2.12)
6
Theoretical Background
where V and I are voltage and current respectively. So, using the Ohm’s law (I = V /R),
power P , dissipated in a wire (or a coil) can be calculated as:
P = I 2R (2.13)
where I and R are the current and resistance of the wire respectively. In case of
magnetic switches, power dissipation comes from a coil, carrying electric current needed
for to generate such magnetic field that the cantilever is attracted, closing the switch
(Figure 1.1). Resistance of the wire R can be calculated using the following formula:
ρL
R= (2.14)
A
where ρ, L and A are wire material’s resistivity, length and cross-section respectively.
The laws of physics are usually expressed in terms of partial differential equations
(PDEs). However, in most cases it cannot be done analytically. Instead, there are
different numerical methods that calculate an approximation of the real solution of the
PDEs. The finite element method (FEM) is one of those [9].
FEM is a powerful tool for solving differential equations. This method can easily
deal with complex geometries and higher-order approximations of the solution [10]. It
revolves around an idea where a function u can be approximated by a function uh using
linear combinations of basis functions [9]:
u ≈ uh (2.15)
X
uh = ξi ϕi (2.16)
i
7
Theoretical Background
This process starts with rewriting the original PDE as its weak form by multiplying
both sides of it by a test function v, and integrating, thus getting the variational
formulation. The test function v must be square integrable and zero at Dirichlet
boundaries [8].
At this stage the equation exists on continuum, the global domain of the prob-
lem. Therefore, the collection of admissible functions and trial functions span infinite-
dimensional functional spaces.
Next step is to discretize the weak form by subdividing its space into smaller sub-
domains or elements, defining the finite element formulation. This step is equivalent to
projection of the weak form of PDEs onto a finite-dimensional subspace. The notations
uh and vh represent the finite-dimensional equivalent of u and v respectively [11]. Since
uh belongs to space of vh , it can be rewritten as the following linear combination:
X
uh = ξi ϕi (2.17)
i
where ϕi and ξi are the basis functions and coefficients to be determined respectively.
By using those basis functions as load functions, the finite element formulation can be
rewritten as:
Aξ = b (2.18)
where A is the stiffness matrix and b is the load vector, both of which can be calculated.
By solving Equation 2.18, it is possible to find ξ, which allows the solution of Equation
2.17.
In essence, FEM represents the original domain of the problem as a collection of
elements. Then, for each element, substitutes the original PDE problem by a set
of simple equations that locally approximate the original equations. After that, it
applies boundary conditions for boundaries of each element and assembles the resulting
equations and boundary conditions into a global system of equations that models the
entire problem. And finally, solves the resulting system of equations [11].
8
Theoretical Background
Microfabrication owes most of its techniques to the semiconductor industry, the same
tools can be used for MEMS fabrication (Figure 2.4). The core process includes cleaning
the substrate, deposing a thin film, applying mask, etching and repeating, if needed
[12].
9
Theoretical Background
10
Theoretical Background
the surface of the silicon substrate using multilayer depositions and patterning of these
thin films (Figure 2.6 (b)) [13].
Figure 2.6: Surface (a) and bulk (b) micromachining processes. Source: [15]
11
3. State of the Art
The past 25 years have seen MEMS transition from being a research curiosity to a
multibillion dollar commercial enterprise. At the same time, there have not been many
commercially successful MEMS, compared to the number of prototype devices created
as part of research. In large part this is because MEMS development is still in an
exploratory phase, where all ideas are considered worth examining [16].
The kingdom of MEMS actuators mainly consists of four families: electrostatic,
piezoelectric, thermal and magnetic [16]. Each actuation principle has its own ad-
vantages and disadvantages. The choice and the optimization of an approach should
be made according to the requirements of a particular application [17]. Electrostat-
ical actuation is the most frequently applied principle combining versatility and simple
technology. It needs neither additional elements like coils or cores, nor special materials
like shape-memory-alloys or piezoelectric ceramics, consumes little power, easily integ-
rated and controlled [18]. On the contrary, the other types of actuators are more robust
and more capable of producing larger forces [17]. Although nearly every permutation
of activation principle and device has been tried, only a few have been leaving the re-
search laboratories [18]. Still, MEMS actuators are used in many fields, such as optics,
medicine, communication systems, automation, aerospace, medicine etc. [17]. Typical
MEMS devices are micropumps, microvalves, microrelays, microgrippers, micromirrors
etc. [18].
In scope of this project I am going to inspect MEMS actuators that are used as
switches. MEMS switches are substantially different from PIN semiconductor diodes
or FET switches, although the purpose of both types of devices is to vary the imped-
ance of an electrical path in a controlled fashion. While solid state devices employ
12
State of the Art
Figure 3.1: Switching principles for ohmic and capacitive devices. Source: [2]
MEMS relays generally offer a number of advantages over solid state devices such
as higher off-state isolation and low power consumption, depending on the actuation
scheme. On the other hand, MEMS switches suffer from a series of problems in terms
of reliability, particularly exacerbated by failure mechanisms such as self-actuation,
stiction, electromigration, microwelding, etc. MEMS devices additionally carry the
burden of needing an appropriate packaging solution that guarantees functionality and
reliability, which potentially increases manufacturing costs. All of these challenges are
currently subjects for process improvements and optimization [2].
As discussed previously, there are four main groups of MEMS actuators. Same can
be applied to MEMS switches. In this work I am going to focus on electromagnetic
MEMS switches.
Magnetic MEMS are based on the interaction between sources of electromagnetic
or magnetic forces such as coils or permanent magnets and microstructures fabricated
with magnetic materials. The strong interest in the application of such old and well-
established physics to microscale components lies in the advantages offered by magnetic
forces over conventional electrostatic components at smaller scales [2].
There is a number of magnetic MEMS switches reported in literature (Table 3.1).
13
State of the Art
One of the first ones dates back to 1993, when Ahn and Allen [19] suggested a mag-
netic microactuator, which would have a magnetic core with a conductor coil wrapped
around it (Figure 3.2). A current through the coil would generate a magnetic flux in
the core and the air gap, generating a magnetic force in the air gap, which would at-
tract the cantilever to the bottom contact. With a cantilever, which was 2.5 µm thick,
25 µm wide and 780 µmlong, they achieved a deflection of 6 µm by applying 800 mA to
the coil, which results in 832 µW power consumption.
14
State of the Art
A few years later, in 1997, Judy and Muller [20] introduced a concept of local
magnetic forces generated by a coil (Figure 3.3). It requires 500 mA to rotate the plate
with magnetic material by 45◦ , which is 450 µm wide and 450 µm long, resulting with
1.6 W power dissipation.
15
State of the Art
Figure 3.4: Concept scheme of a magnetic MEMS switch [22]. Source: [22]
along their easy axis corresponding to their length. A magnetized cantilever experiences
a torque that aligns its magnetization to the field axis and orientation. The device
utilizes the embedded coil to force the cantilever in one of two possible magnetic states,
depending on the polarity of the applied voltage, thus magnetizing the cantilever along
its length in either direction. The bias field provided by the permanent magnet then
exerts a torque on the hinged cantilever, pulling either of its ends towards the bottom
substrate while the other end is pushed upwards. Once the cantilever is forced in a
magnetic state, and is consequently aligned to the bias field, the electrical current can
be turned off, as the magnetization is then induced solely by the bias field.
The bistable latching device offers great performance in terms of power consump-
tion, as only short current pulses are needed to excite the magnetic cantilever. To
change state, it requires 79 mA of current for 0.2 ms, which results in 93 µJ energy
consumption. Low contact resistance values are also measured, around 50 mΩ.
The work reported in 2005 by Cho et al. [24] represents an attempt to integrate
electromagnetic and electrostatic actuation mechanisms in a single device, with the
16
State of the Art
Figure 3.5: Concept scheme of a latching magnetic MEMS switch [23]. Source: [23]
aim of providing latching functionality and ensuring low power consumption. The
device comprises an insulating movable membrane fabricated as a stack of nitride and
a microcoil (Figure 3.6). When the device is immersed in a uniform magnetic field,
it is possible to inject a current in the coil to generate a magnetic moment, which
experiences magnetic forces that align its orientation to the external field lines.
Figure 3.6: Concept scheme of a latching magnetic MEMS switch [24]. Source: [24]
When the suspended membrane is actuated, it approaches the plates and a voltage
can then be applied to exert a force between them. This can act as a latching mechan-
ism that maintains the on state by simply applying a voltage to charge the capacitor.
17
State of the Art
The membrane is 500 µm wide, 700 µm long and 1 µm high, in addition to coil
thickness of 3 µm. To change state the device requires 53 mA, which results in 40.3 µJ
of energy needed. The electrostatic hold voltage is as low as 3.7 V and the contact
resistance was determined to be 0.5 Ω.
Of course, there are many more other designs for magnetic MEMS switches [1, 25,
26], but the ones covered in this chapter represent most of the ideas that currently
exist.
18
4. Concept of a Magnetic MEMS Switch
This project has started with an idea, where a magnetic MEMS swtich would have a
conductive cantilever (polysilicon with Young’s modulus of 162.8 GPa) with a magnetic
film (cobalt with remanent magnetization of 1.75 T) attached to it, while there is planar,
rectangular coil (copper with resistivity of 1.68 × 10−8 Ω m) beneath it, that creates a
magnetic field, attracting the cantilever and closing the switch (Figure 1.1). The cross-
section of the suggested device can be seen below (Figure 4.1).
To simplify the problem, in this work I am only focusing on modelling the deflection
of the cantilever and the magnetic force due to the magnetic field exerted by the coil,
so in Figure 4.2 I show which region of the device is discussed in scope of this project.
Magnetic actuation in this case involves two branches of physics: classical mechan-
ics, which deals with calculation of cantilever deflection due to magnetic forces applied
to the magnetic film, and electromagnetism, which deals with calculation of magnetic
forces due to the magnetic field as well as the calculation of that magnetic field exerted
by the current in the coil.
So it was pretty logical for me to divide this model into two simpler ones, each
dealing only with one branch of physics. The one that calculates the magnetic force
19
Concept of a Magnetic MEMS Switch
exerted by the coil includes the coil itself and the magnetic film (Figure 4.3). This force
is then “fed” to the model that calculates the deflection, which in its turn includes the
cantilever itself and the magnetic film, to which the force is applied (Figure 4.4).
Figure 4.3: Sub-problem that solves the device for magnetic force
At this stage, the dimensions are not defined yet. The only set in stone thing is that
the cantilever beam should not be longer or wider than 70 µm, which is a constraint
that I set for the device, so that in total it is not bigger than ≈ 100 µm.
20
5. Modelling of a Magnetic Actuation
The modelling problem can be divided into two simpler subproblems: modelling of the
coil and magnetic forces and modelling of the cantilever’s deflection.
To solve a problem in COMSOL, usually, there are well-defined steps to do that. First,
I have created a 3D model according to Figure 4.3, although only with one loop of wire
for the sake of simplicity. The dimensions for now were chosen arbitrarily, but within
reason (Figures 5.1).
21
Modelling of a Magnetic Actuation
Then the model (Figure 5.2) was put inside of an air domain, which in its turn is
put inside of an infinite domain to simulate infinite space around the model to “damp”
the magnetic field beyond the air domain (Figure 5.3).
Figure 5.4: Influence of having an infinite domain versus not having one
The model was set up for solving the Maxwell’s equations for stationary problems
using the “Magnetic Fields” interface. The outer boundary of the infinite domain is
magnetically insulated, meaning that the magnetic field does not exit the modelling
domain, but, like it was discussed above, that is not a problem because of how the
22
Modelling of a Magnetic Actuation
infinite domain “damps” the field. The coil is prescribed with some current density,
following its shape in clockwise direction.
The magnetic film’s material is cobalt with remanent magnetization of 1.75 T (Fig-
ure 5.5), pointing in −z direction, so that when the coil has the current flowing in
clockwise direction, like defined above, the film would experience attractive forces.
The model was meshed using normal COMSOL settings, with a bit denser mesh
in and around the magnetic film. Then it was solved using the “Stationary Solver” of
COMSOL and the force acting on the film was calculated to be −3.7 µN.
However, it was not clear if the solution is converged here, so it was necessary to do
the convergence study by solving the same model multiple times, each time reducing
mesh element size. The solution is converged, when reducing mesh size does not affect
the solution and it stabilizes, like it is shown in Figure 5.6. So the final result for this
model configuration is around −4.5 µN.
Solving the cantilever for displacement is a simpler task, as it does not require a medium
to be modelled in. Other than that, the procedure is similar to the one described in
the section above.
23
Modelling of a Magnetic Actuation
Figure 5.6: Convergence study for coil and magnetic film modelling, n is a mesh size
parameter, the bigger it is, the smaller mesh becomes
First, the dimensions for the cantilever beam must be chosen. In Chapter 5.1, I
have set the magnetic film to be 10 µm wide, 10 µm long and 1 µm high. I wanted to
model a very simple rectangular beam, so I chose it to be 10 µm wide as well, 50 µm
long and 0.5 µm high with a magnetic film attached to the top of it (Figure 5.7).
Then I have used the “Solid Mechanics” interface of COMSOL, which is based
on solving the equations of motion to get results such as displacements, stresses and
strains. In that interface, I have applied such boundary conditions for the system as
evenly distributed force on the magnetic film and how the cantilever is fixed, so that
the system would look as in Figure 5.8.
24
Modelling of a Magnetic Actuation
And as a final step, like in Chapter 5.1, I have performed a mesh convergence study,
continuously reducing the mesh size while solving the model, using the “Stationary
Solver”, as there are not any time-dependent variables. And when the solution is
converged, it becomes the final result. For example, in this case, if I apply total force
of 4.5 µN pointing in −z direction, the maximum deflection would be around 9.27 µm
(Figure 5.9).
Figure 5.9: Convergence study for cantilever modelling, n is a mesh size parameter,
the bigger it is, the smaller mesh becomes
25
6. Model Verification
After I have performed the simulations in Chapers 5 and got converged results, I have
decided to compare those results with analytic solutions, using methods discussed in
Chapters 2.1, 2.2 and 2.3, for the sake of checking if the simulations’ results were
correct.
In Chapter 2.1 I have covered two methods of calculating the magnetic force acting
upon a magnet. The second method, which takes use of Maxwell’s stress tensors
(F = µ10
RR
S · n dA) is being used by COMSOL, while I will use
Z
F = ∇(M · B) dV (6.1)
to calculate the force and compare my results with the simulation’s results.
First, I needed to calculate the magnetic field exerted by a square electric loop. To
do that, I had to apply the Biot-Savart’s law (Chapter 2.2) for a straight wire:
I
B= (cos α1 + cos α2 ) (6.2)
4πd
which holds for an infinitely thin wire, which is a reasonable approximation for most
cases. And by splitting the loop into 4 straight fragments, calculating magnetic field
due to each one of them and adding the results together, I came up with a magnetic
field as a function of position relative to the loop (Appendix A.1).
26
Model Verification
For example, if I have a loop that is 10 µm wide and 10 µm long and carries a
current of 1 A (like in Chapter 5.1), the magnetic field in the center would be around
9 × 104 A m−1 (Figure 6.1).
Figure 6.1: Magnetic field in the center of a square electric loop due to 1 A of current
As I am interested only in forces along z axis, Equation 6.1 takes the form of:
Z
∂Bz
Fz = Mz dV. (6.3)
∂z
from which follows that to calculate the magnetic force F in z direction, I need to
take partial derivatives with respect to z of the z components of the magnetic field B,
integrate the results across the volume V and multiply it by magnetization Mz .
Calculation of the derivative and integration have been done numerically in MAT-
LAB, as the expression for the magnetic field became quite lengthy.
So, if I try to replicate the model from Chapter 5.1, assuming thin wires (resulting
with distance between the loop and the magnetic film to be 1.5 µm instead of 1 µm),
then the resulting magnetic force would be −4.58 µN, compared to −4.5 µN, which was
calculated using COMSOL.
From this result, two things can be concluded: that the simulations in COMSOL
were performed correctly and that even though the results were close, the geometry of
27
Model Verification
the wires of the electric loop has its effect, since the results from COMSOL converge a
little bit away from the value that was analytically calculated (Figure 5.6).
d4 y
EI = w(x). (6.4)
dx4
To efficiently model a beam, a load diagram is needed. In Figure 6.2 I assume that
second moment of inertia and the Young’s modulus are constant here, meaning that
the cantilever will be solved without having the magnetic film attached to it.
X
F = 0 = −Fmagnetic + Freaction (6.5)
28
Model Verification
X
M = 0 = rFmagnetic − Mreaction (6.7)
where r = 45 µm. The load function w(x), using the singularity functions, would then
take a form of:
According to Equation 6.4, w(x) has to be integrated 4 times to find the deflection
function y(x):
1 Fmagnetic 3 Freaction 3 Mreaction 2
y(x) = − hx − ri + hx − 0i − hx − 0i . (6.10)
EI 6 6 2
Substituting (6.6) and (6.8):
1 Fmagnetic Fmagnetic rFmagnetic
y(x) = − hx − ri3 + hx − 0i3 − hx − 0i2 (6.11)
EI 6 6 2
which becomes the final expression for deflection y(x).
The material used for the cantilever beam is a polycrystalline silicon (polysilicon),
which has the Young’s modulus of 162.8 GPa and second moment of inertia:
29
7. Modelling Efficiency
Another important topic in this project is how efficiently the problems are solved.
Finding a converged solution of a simulation can take quite a bit of time and computer
memory. Reducing those is crucial if I want to optimize the model, as I will have
to run those simulations a number of times, while changing its parameters (Chapter
reference). And if the model is not solved efficiently, it may take days for it to find
the solutions, provided that a computer does not run out of memory. So, there are
two things I came up with that can be used to improve this situation: symmetry and
analytic solution for the magnetic force.
The first thing I noticed is that if the coil and magnetic film are of rectangular shape
(Figure 5.3), it has two planes of symmetry, meaning that it is possible to model only
a quarter of that model (Figure 7.1) and still get the correct magnetic force just by
multiplying the result by 4.
The boundary condition for the surfaces, where the “cut” was made is Magnetic
Insulation, meaning that the magnetic field is forced to be parallel to the boundary,
while the currents is forced to be perpendicular to the boundary. In essence, it creates
a mirror symmetry plane for the magnetic field (n × A = 0, where n and A are a vector
perpendicular to a surface of a boundary and a magnetic vector potential respectively).
If I try to simulate the model from Chapter 5.1 using the symmetry, then, as a
result, I will get the same magnetic force of −4.5 µN, but 4 times faster (Figure 7.2).
30
Modelling Efficiency
Another idea to improve the calculation speed for the model was to calculate the
magnetic force as in Chapter 6.1, but purely numerically in COMSOL and compare if
this solution converges faster than in Figure 7.2.
R
From Figure 7.3 it can be easily concluded that if I am using F = ∇(M · B) dV
instead of F = µ10
RR
S ·n dA (COMSOL’s method) to calculate the magnetic force, the
results are more stable and converge faster, reducing the simulation time substantially.
As a result, in the future, I will be using the analytic solution method.
Meshing is a process that I have barely talked about in this work, even though it is
one of the most important things during FEM. Especially in this work, it is crucial to
set up the mesh properly and effectively.
For simple geometries, like for a cantilever, the default COMSOL settings work
just fine, producing pretty accurate results even without mesh refinement. However,
when modelling a coil and a magnetic film, the model becomes much more complex
31
Modelling Efficiency
(Figure 7.4). It would include an air sphere, a bigger sphere of air with infinite domain
properties, a coil and a magnetic film. It results with a model of a relatively big size.
And if I wanted to do a convergence study by refining mesh, the computation time
could grow out of control very fast. So it was necessary to determine which regions
were required to be refined in that study, and which not.
For example, the infinite domain layer should just have around 5 elements in its
thickness to work as intended, while a magnetic film would have to have a very dense
mesh, since a gradient of a magnetic field must be taken inside of it and having a coarse
mesh there would severely affect the results. An electric coil, in its turn, does not
require a dense mesh, since not much is calculated inside of it, because it is prescribed
with current density already, so it basically acts as a boundary condition. And lastly,
an air domain also does not require a dense mesh, since a magnetic field approximations
are very well done and the geometries are not complex at all.
Of course, convergence studies must be performed either way, but deciding which
mesh regions should be dense right away in advance, definitely helps with speeding up
the simulations, as it eliminates the condition when the program refines mesh across
the whole geometry, while only one piece of it actually needs it.
32
Modelling Efficiency
33
8. Design and Optimization of a Magnetic
MEMS Switch
In this chapter I will describe how I performed the design and optimization of the
switch, starting with defining fabrication constraints, then specifying dimensions for
the two proposed concepts, running the simulations to find the best set of parameters
and showing the results.
One of the goals of this project is to design a magnetic switch, which is viable in terms of
fabrication in Fraunhofer ISIT. To determine the possibilities of their technologies, me
and Roana arranged a meeting with Fabian Lofink, who is responsible for fabrication
of microcantilevers at Fraunhofer ISIT. With his insight, we were able to determine
minimum and maximum dimensions for the switch that can be manufactured, as well
as some other properties (Table B.1).
34
Design and Optimization of a Magnetic MEMS Switch
Using the information from Chapter 8.1, I was able to determine most of the dimensions
for a minimized magnetic switch, which would have to consume the least amount of
power. The sketch for the overall switch setup has already been developed in Chapter
4 (Figure 4.1).
The maximum length and width for a cantilever were decided to be 70 µm, since I
have wanted the whole device to be less than 100 µm wide and 100 µm long. Hence,
the cantilever length becomes 70 µm and its height — 0.5 µm, as it is most beneficial
to have it as long as possible, since that will allow for easier deflection, resulting in
smaller power consumption. The gap between the cantilever is also 0.5 µm, hence the
distance between a coil and a magnetic film is 1.5 µm. This distance is also required
to be minimized, since the closer the magnetic film is to the coil, the stronger forces
35
Design and Optimization of a Magnetic MEMS Switch
would be generated. Going for the smallest possible gap is also essential, as the smaller
it is, the less force is required to close the contact. The side dimensions can be seen
on Figure 8.1. But when it comes to deciding how wide the cantilever should be, two
different concepts arise. Concept 1 states that its width should be as small as possible
— 1.5 µm (Figure 8.2). Concept 2 states that the total width of the cantilever should
be 70 µm, but with having two arms, 1.5 µm each (Figure 8.3).
Concept 1 is easier to fabricate and more compact, but Concept 2 will allow for
easier deflection and hence smaller power consumption. The coil’s inner width in each
concept should be calculated according to Appendix B.3:
af ilm − d(l − 1) − wl,
if l = 1, 3
ain = (8.1)
af ilm − d(l − 2) − w(l − 1), if l = 2
where af ilm , d, l, w are width of the magnetic film, distance between coil loops, number
of coil loops and width of the wires in the coil. The distance between coil loops d is
1.5 µm, as dictated by the fabrication constraints, providing that the best performance
36
Design and Optimization of a Magnetic MEMS Switch
is required. Note that minimum inner width of the coil ain cannot be less than d, so
if it becomes less than d, when being calculated according to (8.1), then it is set to be
equal to d — 1.5 µm. Similar equation has been derived for the inner length of the coil:
bf ilm − d(l − 1) − wl,
if l = 1, 3
bin = (8.2)
bf ilm − d(l − 2) − w(l − 1), if l = 2
where bf ilm is the magnetic film’s length. And like with ain , bin cannot be smaller than
d, so if it is, after calculating it with (8.2), it becomes equal to d, which in this case
would be 1.5 µm.
Still, there are some parameters left undefined. For example, it is unknown what
is more beneficial, to have a larger cross-section of the wires in the coil to get lower
resistance, or to have them smaller, to achieve a more “concentrated” force from them.
The optimal number of loops is also unknown, as well as the optimal height (hf ilm ) and
length of the magnetic film, as increasing those would result in a more stiff cantilever.
The width of the magnetic film (and the cantilever, as they are equal), in its turn,
depends on the concept (1.5 µm for Concept 1 and 70 µm for Concept 2).
To determine which configuration of parameters gives the best results, I have conducted
a series of simulations in COMSOL.
The simulations were performed both for Concept 1 and Concept 2, while solving
for all combinations of parameters (from their minimum values to their maxumum),
that were left undefined (Table 8.2). The methodology of modelling this problem is
the same as described in Chapter 5 - dividing the system into 2 subsystems, but this
time using more effective approaches as described in Chapter 7.
In a model that solved an electric coil for a magnetic force, which it exerts, para-
meter bf ilm was swept through only 2 values, while a model that solved a cantilever
37
Design and Optimization of a Magnetic MEMS Switch
for its deflection, has swept this parameter through 90 values. The reason for that is
because when solving the coil, the magnetic force is linearly dependent on the magnetic
film length, assuming that the coil’s inner length changes as well, according to (8.2).
This relation has been shown in Appendix B.2.
Another thing that is worth covering is the minimal value of bf ilm , which is 25.5 µm,
which is vastly different from 1.5 µm, like the constraints allow. This was done in order
to satisfy equation (8.2), so that bin ≥ d for all parameters. And like the results will
show later, the optimal bf ilm is bigger than 25.5 µm anyway.
For each concept, a mesh convergence study was performed only once for the fol-
lowing parameter values: w = 1.5 µm, h = 0.5 µm, bf ilm = 70 µm, hf ilm = 1 µm, l = 1,
which would give the thinnest and most prolonged model. And the mesh element sizes,
which were determined in this study that lead to a converged result, were used for sim-
ulating other parameter configurations, since I have assumed that the required mesh
element size will not be smaller in other simulations.
Like I have already mentioned, simulation of each configuration of a magnetic switch
is split into two models: one that solves a cantilever beam for the deflection due to
some force, and one that solves a coil for the magnetic force that it exerts upon a
magnetic film due to some current. Force and current in those respective simulations
are arbitrary, in this case: 10 nN and 1 A. The goal of those simulations is to find a
relation between how much current is applied and how much magnetic force is generated
38
Design and Optimization of a Magnetic MEMS Switch
as well as between how much force is applied to a cantilever and how big deflection
it causes. But those relations are linear (Equations (2.5) and (2.9)), so it allows me
to extrapolate those results to see how much force is needed to achieve a maximum
deflection of 0.5 µm, and knowing the required force, it also gives me the information
of how much current is needed to achieve this force.
For example, if a force of 10 nN gives a deflection of 1 µm and a current of 1 A
gives a force of 1 µN, then the required force to achieve 0.5 µm deflection is 5 nN, and
consequently, the required current becomes 0.005 A.
Using this principle, I have calculated the required current to actuate a switch
(reach a deflection of 0.5 µm), which also gives the information of required power to
do that. And by comparing the required power for each configuration of parameters, I
was able to determine which one is the best in terms of power consumption. MATLAB
script that processes the results from simulations the way it was described above, can
be found in Appendix C.2.
The results from simulations have been compiled in a number of plots, for Concept
1 (Figure 8.7) and Concept 2 (Figure 8.8). Although they may not provide accurate
data, they clearly show, which configuration of parameters gives the best performance
in terms of power consumption. Of course, power here is a function of all parameters,
however on those figures, its dependency on bf ilm is not shown. In essence, each data
point there should also be a function of bf ilm , but instead, the heat maps here already
show the global minimum of those functions with a corresponding bf ilm value with it.
From Figure 8.7 it can be seen that the best results for Concept 1 are when l =
1 and hf ilm = 3 µm. After a closer inspection of that plot (Figure 8.4), the least
amount of power is consumed when w = 5.5 µm and h = 3 µm. This configuration
of parameters results with 4.25 µW power consumption when bf ilm = 30 µm and the
39
Design and Optimization of a Magnetic MEMS Switch
current is 13.8 mA, which also is a global minimum of the system, since it may only
be smaller when h > 3 µm or when hf ilm > 3 µm, but that is not possible because
of fabrication constraints at Fraunhofer ISIT. But in general, increasing hf ilm and h
benefits the system.
Results of simulations for Concept 2 are presented on Figure 8.8. A plot with
the results when l = 1 and hf ilm = 3 µm shows the best performance in terms of
power consumption, however upon closer inspection of that plot, it can be seen that a
better configuration may be found by increasing w. After running the simulations for
Concept 2 again, with l = 1 and hf ilm = 3 µm, but now sweeping w from 4 µm to 11 µm
instead (Figure 8.6), the global minimum of power has been found when w = 10 µm
and h = 3 µm, which is 180 nW when bf ilm = 25.5 µm and the current is 1.3 mA.
Figure 8.4: Simulation results of Concept 1 for different parameter configurations when
l = 1 and hf ilm = 3 µm
40
Design and Optimization of a Magnetic MEMS Switch
Figure 8.5: Simulation results of Concept 2 for different parameter configurations when
l = 1 and hf ilm = 3 µm
Figure 8.6: Expanded simulation results of Concept 2 for different parameter config-
urations when l = 1 and hf ilm = 3 µm
41
Figure 8.7: Simulation results of Concept 1 for different parameter configurations
42
Design and Optimization of a Magnetic MEMS Switch
Figure 8.8: Simulation results of Concept 2 for different parameter configurations
43
Design and Optimization of a Magnetic MEMS Switch
9. Conclusion and Future Work
As a result of this work I have successfully designed and modelled two concepts of
a magnetically actuated MEMS switch, which can be fabricated at Fraunhofer ISIT.
Their designs have been minimized for required actuation power. Concept 1 requires
4.25 µW of power and 13.8 mA of current to actuate, while Concept 2 requires just
180 nW of power and 1.3 mA of current. Even though the difference in power between
these two concepts is substantial, it is not quite clear which one is better, since Concept
1, in its turn, is about 46 times narrower than the other one, so that would be more of
a question of application and additional constraints.
Actuation power, current and dimensions are not the only parameters that describe
a magnetic MEMS switch. Some other important ones are on and off resistances,
switching speed, lifetime, contact force, maximum frequency. So, in the future, more
simulations have to be performed. For example, a frequency analysis of the structure
has to be done to calculate resonant frequencies. In addition to that, it also must be
calculated how much current can the switch carry. This is needed to fully characterize
the system, so that it can be fairly compared to others that are currently on the market.
This work has also covered some of the research performed in this direction, theor-
etical background of magnetic actuation and the process of modelling such systems in
COMSOL, combining both numerical methods and analytical ones, as well as efficiency
improvements of those simulations. Also, it includes some additional simulations that
examine influence of deformation of a square-shaped coil on force that it exerts and
show that deforming the wire does increase the force and, in fact, is more power efficient
than if this force is increased by adding more current.
44
Bibliography
[6] J. D. Jackson, Classical Electrodynamics. John Wiley & Sons, Inc., 1962.
[8] M. G. Larson and F. Bengzon, The Finite Element Method: Theory, Implement-
ation, and Practice. Springer, 2010.
[9] COMSOL. (n.d.). Detailed explanation of the finite element method (fem), [On-
line]. Available: https://www.comsol.dk/multiphysics/finite- element-
method.
45
Bibliography
[11] MathWorks. (n.d.). Finite element method (fem) basics, [Online]. Available:
http://se.mathworks.com/help/pde/ug/basics-of-the-finite-element-
method.html.
[14] K. F. Lei, “Chapter 1: Materials and fabrication techniques for nano- and micro-
fluidic devices”, in Microfluidics in Detection Science: Lab-on-a-chip Technolo-
gies, The Royal Society of Chemistry, 2015, pp. 1–28.
[17] H. Fujita, “Microactuators and micromachines”, Proceedings of the IEEE, vol. 86,
pp. 1721–1732, 8 1998.
46
Bibliography
[21] J. A. Wright, Y.-C. Tai and S.-C. Chang, “A large-force, fully-integrated mems
magnetic actuator”, in Proceedings of International Solid State Sensors and Ac-
tuators Conference (Transducers ’97), vol. 2, 1997, pp. 793–796.
[24] I.-J. Cho, T. Song, S.-H. Baek and E. Yoon, “A low-voltage and low-power rf
mems series and shunt switches actuated by combination of electromagnetic and
electrostatic forces”, IEEE Transactions on Microwave Theory and Techniques,
vol. 53, pp. 2450–2457, 7 2005.
[26] S. Lu, T.-H. Lin, S. Paul and H. Lu, “A study on the performance and reliability
of magnetostatic actuated rf mems switches”, Microelectronics Reliability, vol. 49,
pp. 59–65, 1 2009.
47
A. Calculations and Derivations
µ0 Idl × ~x
dB = (A.1)
4π |~x|3
where dB, µ0 , I, dl and ~x are elemental flux density, current, vacuum permeability
(4π × 10−7 ), element of length and position vector respectively (Figure A.1).
Using (A.1) we can calculate magnetic fields due to a straight current wire [4]
µ0 I
B= (cos α1 + cos α2 ) (A.2)
4πd
where d, α1 and α2 are distance to the wire and angles that are shown on Figure A.2
respectively. This equation can be expanded to calculate a magnetic field exerted by a
square loop simply by splitting the loop into 4 straight fragments, calculating magnetic
field for each one of them and adding the results together.
48
Calculations and Derivations
Figure A.2: Magnetic field B at point P due to an electric current I in a straight wire
!
µ0 I 1 a1 a2
B1 = p p +p 2 . (A.3)
4π b21 + z 2 b21 + z 2 + a21 b1 + z 2 + a22
Its x, y and z components are then:
49
Calculations and Derivations
z b1
B1x = 0, B1y = −B1 p 2 , B1z = B1 p 2 . (A.4)
b1 + z 2 b1 + z 2
Magnetic field at point P due to wires 2, 3 and 4 are calculated analogously:
!
µ0 I 1 b1 b2
B2 = p p +p 2 (A.5)
4π a22 + z 2 a22 + z 2 + b21 a2 + z 2 + b22
z a2
B2x = B2 p 2 , B2y = 0, B2z = B2 p 2 (A.6)
a2 + z 2 a2 + z 2
!
µ0 I 1 a1 a2
B3 = p p +p 2 (A.7)
4π b2 + z 2
2
b22 + z 2 + a21 b2 + z 2 + a22
z b2
B3x = 0, B3y = B3 p 2 , B3z = B3 p 2 (A.8)
b2 + z 2 b2 + z 2
!
µ0 I 1 b1 b2
B4 = p p +p 2 (A.9)
4π a21 + z 2 a21 + z 2 + b21 a1 + z 2 + b22
z a1
B4x = −B4 p 2 , B4y = 0, B4z = B4 p 2 . (A.10)
a1 + z 2 a1 + z 2
To get the total magnetic field at point P , all magnetic fields due to wires 1, 2, 3
and 4 must be added together:
Btotal = B1 + B2 + B3 + B4 . (A.11)
Same thing can be done if only specific components of the magnetic field are re-
quired:
50
Calculations and Derivations
A magnetic force exerted by a magnetic field on a single moving charge is defined by:
dF = qv × B (A.15)
where q is charge, v its velocity and B is external magnetic field. It can be rewritten
as:
dF = I(dl × B) (A.16)
where I is electric current flowing through a wire segment dl. The total force on the
whole wire would be:
Z
F =I (dl × B) (A.17)
which is called the Laplace force. This expression can be written as an integral over a
surface bounded by the wire:
I
F =I (dl × B). (A.18)
H
Using Stokes’ theorem, dl is replaced by the operator dµ × ∇, obtaining dl × B =
R
(dµ × ∇) × B. Now:
Z
F =I (dµ · ∇)B. (A.20)
51
Calculations and Derivations
Z
m=I dµ (A.21)
with dµ = 21 r ×dl, where r is the position vector. The product dµ is equal in magnitude
to the triangular area formed by the vectors r and dl.
Assuming that B is almost uniform, I can take it out of the integral in (A.20), and
after substituting (A.21) into (A.20), I get:
F = (m · ∇)B. (A.22)
Since m is constant:
But in (A.23) I calculate the force acting only on a single loop and I need to find
force acting on a magnet with volume V and magnetization M . If I assume that magnet
is a collection of infinitely small current loops, then using (A.23), it can be said that
the magnetic force would be:
Z
Ftotal = ∇(M · B) dV. (A.24)
d2 d2 y
EI 2 = w(x) (A.25)
dx2 dx
where E, I, y, x and w are the Young’s modulus, second moment of inertia, deflection,
position and load respectively.
According to Figure A.4, load w is expressed as:
52
Calculations and Derivations
where r = 45 µm. Calculating shear force V , bending moment M and deflection y(x):
Z x
− V (x) = w(x) dx = −Fmagnetic hx − ri0 +
−∞
Z x
M (x) = − V (x) dx = −Fmagnetic hx − ri1 +
−∞
Z x
dy(x) Fmagnetic
EI = M (x) dx = − hx − ri2 +
dx −∞ 2
Freaction
hx − 0i2 + Mreaction hx − 0i1 + a (A.29)
2
ZZ x
Fmagnetic
EIy(x) = M (x) dx2 = − hx − ri3 +
−∞ 6
Freaction Mreaction
hx − 0i3 + hx − 0i2 + ax + b. (A.30)
6 2
53
Calculations and Derivations
dy(0)
Applying boundary conditions y(0) = 0 and dx
= 0 leads to:
a = 0, b = 0. (A.31)
X
F = 0 = −Fmagnetic + Freaction (A.32)
X
M = 0 = rFmagnetic − Mreaction (A.34)
1 Fmagnetic 3 Fmagnetic 3 rFmagnetic 2
y(x) = − hx − ri + hx − 0i + hx − 0i . (A.36)
EI 6 6 2
If the material of the cantilever is a polysilicon, its Young’s modulus E is 162.8 GPa
[27]. The cantilever’s cross-section is a rectangle, then its second moment of inertia is:
54
B. Additional Simulations
Concept 1 Concept 2
Cantilever’s length 70 µm 70 µm
Cantilever’s width 1.5 µm 70 µm
Cantilever’s height 0.5 µm 0.5 µm
Magnetic film’s length 30 µm 25.5 µm
Magnetic film’s height 3 µm 3 µm
The material for the cantilever is polycrystalline silicon, which has an elastic limit
of up to 1.2 GPa [27], so the maximum stress must be below that number. After
simulations of Concept 1 (Figure B.1) and Concept 2 (Figure B.2), the maximum
stresses are 15.3 MPa and 14.5 MPa respectively, showing that the structure will not
be permanently deformed during operation. However, how that deflection affects the
magnetic film is unknown, since there is no data on stress-strain behavior for cobalt.
55
Additional Simulations
netic Force
There is a linear relation between magnetic film’s length and magnetic force that is
generated, assuming that the coil’s inner length is changing with the magnetic film’s
length linearly, for example, according to Equation 8.2. On Figure B.3 this linear
relation can be easily seen. It has been modelled with MATLAB script on basis of
Appendix C.1. The setup of this system is of little interest, this simulation just shows
the dynamics of the system while changing one of its parameters. The magnetic film’s
56
Additional Simulations
length has been swept from 25.5 µm to 70 µm with 0.5 µm step size. This relation has
allowed
Figure B.3: Relation between length of a rectangular loop and force that it exerts
To decrease the number of parameters, for which the system should be modelled, it is
necessary to determine if some of them can be expressed in terms of others. In this
case, there must be a relation between the magnetic film’s position and size and coil’s
position and size, which gives the best performance. In case when the coil is infinitely
thin, it is best when the coil replicates the shape of the magnetic film, because there will
no be any counteracting vertical magnetic forces acting on the film. But in this case,
the wires in the coil have a non-zero width which also comes with distance between
loops in the coil, so it was necessary to determine the new rule of coil positioning for
a more complex situation.
Three simulations were performed in COMSOL. In the first simulation (Figure B.4),
the coil had 1 loop, the magnetic film was 40 µm wide and 40 µm long, while its inner
length and width have been swept from 30 µm to 40 µm. Width of the wire is 5 µm.
Both, the coil and the magnetic film are centered. The biggest force is generated when
the coil’s inner length and width are between 35 µm and 36 µm.
57
Additional Simulations
Figure B.4: Relation between coil dimensions and magnetic force that it exerts upon
constant magnetic film. The coil has 1 loop
In the second simulation (Figure B.5), the setup is the same, except that the number
of loops in the coil is 2 and distance between loops is 1.5 µm. Coil’s inner length and
width have been swept from 19.5 µm to 37.5 µm. The biggest force is generated when
the coil’s inner length and width are between 34.5 µm and 35.5 µm.
In the third simulation (Figure B.6), the number of loops in the coil is 3. Coil’s
inner length and width have been swept from 7 µm to 37 µm. The biggest force is
generated when the coil’s inner length is between 22 µm and 23 µm.
Ideally, it is necessary to find the best dimensions for the coil for each parameter
configuration, but to simplify the problem, two equations can be derived, which calcu-
late a very good approximation of the best dimensions of the coil for the corresponding
magnetic film:
af ilm − d(l − 1) − wl,
if l = 1, 3
ain = (B.1)
af ilm − d(l − 2) − w(l − 1), if l = 2
bf ilm − d(l − 1) − wl,
if l = 1, 3
bin = (B.2)
bf ilm − d(l − 2) − w(l − 1), if l = 2
58
Additional Simulations
Figure B.5: Relation between coil dimensions and magnetic force that it exerts upon
constant magnetic film. The coil has 2 loops
where af ilm , bf ilm , d, l, w, ain , bin are width and length of the magnetic film, distance
between coil loops, number of coil loops, width of the wires in the coil, coil’s inner
width and length respectively. For example, if the magnetic film is 40 µm wide and
long, the width of the wires in the coil is 5 µm, there are 2 loops in the coil and distance
between loops is 1.5 µm, then the coil’s inner width and length would be 35 µm, which
is a pretty good approximation, taking into account that these formulas are meant
many different parameter configurations. They can also be valid for all odd or even
loop numbers respectively, but it has not been checked.
In this work I have only focused on designing and modelling rectangular-shaped coils,
but I have also conducted a small study to discover how deforming a rectangular coil
may influence the magnetic force that it exerts upon a magnetic film.
Figure B.7 presents an initially rectangular coil that has been deformed. A number
of simulations were performed, sweeping through offset parameter from 1 µm to 4 µm
59
Additional Simulations
Figure B.6: Relation between coil dimensions and magnetic force that it exerts upon
constant magnetic film. The coil has 3 loops
(Figure B.8). It must be noted that the magnetic film’s shape follows the coil’s shape.
As it can be seen, the generated force increases the more the shape is deformed, keeping
total area constant.
However, increasing offset also increases the power dissipation in the wire, so it
was necessary to see if deformation is more efficient than just increasing the current in
the wire. On Figure B.9 two graphs can be seen: one shows the results of deforming
60
Additional Simulations
the wire, another one shows the results of increasing current of the wire before it was
deformed. In both cases the power dissipation are equal, which gives that deforming
the wire in that way is more efficient, since it gives a larger force.
Even though tweaking coil shapes may result in a better performance, like it was just
shown, this concept has not been further investigated in this work, since it dramatically
complicates modelling, characterization , optimization and fabrication of the system
and I my opinion should be a research on its own.
61
C. MATLAB Scripts
1 close all ;
2 clear all ;
3 clc ;
4
5 I = −1; % c u r r e n t
6 Mz = −1.75; % m a g n e t i z a t i o n o f t h e f i l m
7
8 l e n g t h l = 10 e −6; % width o f t h e l o o p
9 w i d t h l = 10 e −6; % l e n g t h o f t h e l o o p
10
11 l e n g t h f = 10 e −6; % l e n g t h o f t he f i l m
12 w i d t h f = 10 e −6; % width o f t h e f i l m
13 h e i g h t f = 1e −6; % h e i g h t o f t h e f i l m
14
17 steps = 101;
18
19 Bz = z e r o s ( s t e p s , s t e p s , s t e p s ) ; % magnetic f i e l d , z component
20 Bx = z e r o s ( s t e p s , s t e p s , s t e p s ) ; % magnetic f i e l d , x component
21 By = z e r o s ( s t e p s , s t e p s , s t e p s ) ; % magnetic f i e l d , y component
62
MATLAB Scripts
22 normB = z e r o s ( s t e p s , 1 ) ;
23
24 zz = 1 ;
25 f o r z = d : h e i g h t f / ( s t e p s −1) : d+h e i g h t f
26 xx = 1 ;
27 f o r x = l e n g t h l /2− l e n g t h f / 2 : l e n g t h f / ( s t e p s −1) : l e n g t h l
/2+ l e n g t h f /2
28 yy = 1 ;
29 f o r y = w i d t h l /2− w i d t h f / 2 : w i d t h f / ( s t e p s −1) : w i d t h l
/2+ w i d t h f /2
30
31 a1 = l e n g t h l −x ;
32 a2 = x ;
33
34 b1 = y ;
35 b2 = w i d t h l −y ;
36
37 r 1 = s q r t ( a1ˆ2+b1ˆ2+z ˆ2 ) ;
38 r 2 = s q r t ( a1ˆ2+b2ˆ2+z ˆ2 ) ;
39 r 3 = s q r t ( a2ˆ2+b1ˆ2+z ˆ2 ) ;
40 r 4 = s q r t ( a2ˆ2+b2ˆ2+z ˆ2 ) ;
41
63
MATLAB Scripts
r 2 ) −1/(b2ˆ2+z ˆ 2) ∗( a1 / r 3+a2 / r 4 ) ) ;
45
46 yy = yy +1;
47 end
48 xx = xx +1;
49 end
50 normB ( z z ) = s q r t ( Bz ( 5 1 , 5 1 , z z ) ˆ2+By ( 5 1 , 5 1 , z z ) ˆ2+Bx ( 5 1 , 5 1 , z z
) ˆ 2) ; % magnetic f i e l d , norm
51 z z = z z +1;
52 end
53
54 gradM = z e r o s ( s t e p s , s t e p s , s t e p s ) ; % g r a d i e n t o f t h e magnetic
f i e l d , z component
55
56 f o r xx = 1 : s i z e ( Bz , 1 )
57 f o r yy = 1 : s i z e ( Bz , 2 )
58 f o r z z = 1 : s i z e ( Bz , 3 )
59 temp ( z z ) = Bz ( xx , yy , z z ) ;
60 end
61 gradM ( xx , yy , : ) = g r a d i e n t ( temp ) / ( h e i g h t f / s t e p s ) ;
62 end
63 end
64
1 clear all ;
64
MATLAB Scripts
2 close all ;
3 clc ;
4
8 c = 1;
9 optimal model = zeros (9 ,6) ;
10
11 for i d l = 1:3
12 for id h = 1:3
13
14 id t = ’1 ’ ;
15
16 %% r e a d i n g data from c a n t i l e v e r s i m u l a t i o n
17
18 fileName = fopen ( s t r c a t ( ’ c a n t i l e v e r t y p e ’ , id t , ’ ’ , i n t 2 s t r (
id h ) , ’ . txt ’ ) , ’ r ’ ) ;
19 formatSpec = ’%f ’ ;
20 f i l e = f s c a n f ( fileName , formatSpec ) ;
21 d i s p l a c e m e n t = z e r o s ( l e n g t h ( f i l e ) / 2 , 2 ) ; % data from t he
cantilever simulation
22
23 k = 1;
24 for i = 1: length ( displacement )
25 d i s p l a c e m e n t ( i , 1 ) = f i l e ( k ) ; k = k+1; % l e n g t h o f th e
magnetic f i l m
65
MATLAB Scripts
26 d i s p l a c e m e n t ( i , 2 ) = f i l e ( k ) ; k = k+1; % maximum d e f l e c t i o n
due t o 1e−8 N
27 end
28
29 f o r c e r e q u i r e d = z e r o s ( length ( f i l e ) /2 ,2) ; % f o r c e r e q u i r e d to
a c h i e v e 0 . 5 e−6 m d e f l e c t i o n
30 f o r c e r e q u i r e d ( 1 : length ( f o r c e r e q u i r e d ) ,1) = displacement ( 1 :
length ( displacement ) ,1) ;
31
36 %% r e a d i n g data from c o i l s i m u l a t i o n
37
38 fileName = fopen ( s t r c a t ( ’ c o i l t y p e ’ , id t , ’ ’ , i n t 2 s t r ( id h ) , ’
f o r c e l o o p ’ , i n t 2 s t r ( i d l ) , ’ . txt ’ ) , ’ r ’ ) ;
39 formatSpec = ’%f ’ ;
40 f i l e = f s c a n f ( fileName , formatSpec ) ;
41 c o i l d a t a = z e r o s ( l e n g t h ( f i l e ) / 4 , 5 ) ; % data from t he c o i l and
magnetic f i l m s i m u l a t i o n
42 k = 1;
43
66
MATLAB Scripts
47 c o i l d a t a ( i , 3 ) = f i l e ( k ) ; k = k+1; % h e i g h t o f t he w i r e
48 c o i l d a t a ( i , 4 ) = f i l e ( k ) ; k = k+1; % magnetic f o r c e due t o
1 A of current
49 end
50
51 fileName = fopen ( s t r c a t ( ’ c o i l t y p e ’ , id t , ’ ’ , i n t 2 s t r ( id h ) , ’
power loop ’ , i n t 2 s t r ( i d l ) , ’ . txt ’ ) , ’ r ’ ) ;
52 formatSpec = ’%f ’ ;
53 f i l e = f s c a n f ( fileName , formatSpec ) ;
54 k = 0;
55
61 %% i n t e r p o l a t i n g
62
63 variations = 72;
64
68 for l = 1: variations
69 k = l;
70 for i = 1: size ( coil data ,1) / variations
71 int temp ( i , 1 ) = c o i l d a t a (k , 1 ) ;
72 int temp ( i , 2 ) = c o i l d a t a (k , 4 ) ;
67
MATLAB Scripts
73 int temp ( i , 3 ) = c o i l d a t a (k , 5 ) ;
74 k = k+v a r i a t i o n s ;
75 end
76 i n t a r r a y ( : , l , 1 ) = interp1 ( int temp ( : , 1 ) , int temp ( : , 2 ) ,
displacement ( : , 1 ) ) ;
77 i n t a r r a y ( : , l , 2 ) = interp1 ( int temp ( : , 1 ) , int temp ( : , 3 ) ,
displacement ( : , 1 ) ) ;
78 end
79
80 %% g a i n c a l c u l a t i o n
81
82 gain = zeros ( s i z e ( f o r c e r e q u i r e d , 1 ) , v a r i a t i o n s ) ;
83
84 for i = 1: variations
85 gain ( : , i ) = f o r c e r e q u i r e d ( : , 2 ) ./ i n t a r r a y ( : , i , 1 ) ;
86 end
87
88 %% r e q u i r e d power f o r 0 . 5 e−6 m d e f l e c t i o n
89
90 power = i n t a r r a y ( : , : , 2 ) . ∗ ( g a i n . ∗ g a i n ) ;
91
92 %% r e q u i r e d f o r c e f o r 0 . 5 e−6 m d e f l e c t i o n
93
94 f o r c e = i n t a r r a y ( : , : , 1 ) . ∗ ( gain ) ;
95
96 %% f i n d i n g o p t i m a l model
97
98 optimum = z e r o s ( v a r i a t i o n s , 6 , 9 ) ; % c o l l e c t i o n o f
68
MATLAB Scripts
c o n f i g u r a t i o n s , o p t i m i z e d f o r power
99
113 c = c +1;
114
115 end
116 end
69