Wafer-level heterogeneous integration of MEMS actuators

Wafer-level heterogeneousintegration of MEMS actuators

Stefan Braun

MICROSYSTEM TECHNOLOGY LABORATORYSCHOOL OF ELECTRICAL ENGINEERING

ROYAL INSTITUTE OF TECHNOLOGY

ISBN 978-91-7415-493-1ISSN 1653-5146

TRITA-EE 2010:002

Submitted to the School of Electrical EngineeringKTH—Royal Institute of Technology, Stockholm, Sweden,

in partial fulfillment of the requirements for the degree of Doctor of PhilosophyStockholm 2010

ii Wafer-level heterogeneous integration of MEMS actuators

Copyright © 2010 by Stefan Braun

All rights reserved to the summary part of this thesis, including all pictures andfigures. No part of this publication may be reproduced or transmitted in any form orby any means, without prior permission in writing from the copyright holder.

The copyrights for the appended journal papers belong to the publishing houses ofthe journals concerned. The copyrights for the appended manuscripts belong to theirauthors.

Printed by Universitetsservice US AB, Stockholm 2010.

Thesis for the degree of Doctor of Philosophy at the Royal Institute of Technology,Stockholm, Sweden, 2010.

ABSTRACT iii

AbstractThis thesis presents methods for the wafer-level integration of shape memory alloy(SMA) and electrostatic actuators to functionalize MEMS devices. The integrationmethods are based on heterogeneous integration, which is the integration of differentmaterials and technologies. Background information about the actuators and theintegration method is provided.

SMA microactuators offer the highest work density of all MEMS actuators, how-ever, they are not yet a standard MEMS material, partially due to the lack of properwafer-level integration methods. This thesis presents methods for the wafer-level het-erogeneous integration of bulk SMA sheets and wires with silicon microstructures.First concepts and experiments are presented for integrating SMA actuators withknife gate microvalves, which are introduced in this thesis. These microvalves featurea gate moving out-of-plane to regulate a gas flow and first measurements indicate out-standing pneumatic performance in relation to the consumed silicon footprint area.This part of the work also includes a novel technique for the footprint and thicknessindependent selective release of Au-Si eutectically bonded microstructures based onlocalized electrochemical etching.

Electrostatic actuators are presented to functionalize MEMS crossbar switches,which are intended for the automated reconfiguration of copper-wire telecommunica-tion networks and must allow to interconnect a number of input lines to a numberof output lines in any combination desired. Following the concepts of heterogeneousintegration, the device is divided into two parts which are fabricated separately andthen assembled. One part contains an array of double-pole single-throw S-shaped ac-tuator MEMS switches. The other part contains a signal line routing network whichis interconnected by the switches after assembly of the two parts. The assembly isbased on patterned adhesive wafer bonding and results in wafer-level encapsulationof the switch array. During operation, the switches in these arrays must be individ-ually addressable. Instead of controlling each element with individual control lines,this thesis investigates a row/column addressing scheme to individually pull in or pullout single electrostatic actuators in the array with maximum operational reliability,determined by the statistical parameters of the pull-in and pull-out characteristics ofthe actuators.

Keywords: Microelectromechanical systems, MEMS, silicon, wafer-level, inte-gration, heterogeneous integration, transfer integration, packaging, assembly, waferbonding, adhesive bonding, eutectic bonding, release etching, electrochemical etch-ing, microvalves, microactuator, Shape Memory Alloy, SMA, NITINOL, TiNi, NiTi,cold-state reset, bias spring, stress layers, crossbar switch, routing, switch, switcharray, electrostatic actuator, S-shaped actuator, zipper actuator, addressing, transferstamping, blue tape

Stefan Braun, [email protected] Technology Laboratory, School of Electrical EngineeringKTH—Royal Institute of Technology, SE-100 44 Stockholm, Sweden

iv Wafer-level heterogeneous integration of MEMS actuators

... I have not filled this volume with pompous rhetoric, with bombast andmagnificent words, or with the unnecessary artice with which so manywriters gild their work. I wanted nothing extranous to ornament mywriting, for it has been my purpose that only the range of material andthe gravity of the subject should make it pleasing. ...

From “Il Principe”Niccolò Machiavelli, 1469-1532

Translated by Peter Constantine, “The Prince - A new translation”

ABSTRACT v

To my family

vi Wafer-level heterogeneous integration of MEMS actuators

CONTENTS vii

Contents

Abstract iii

List of papers ix

1 Introduction and structure 1

2 MEMS actuators 32.1 Common microactuation mechanisms . . . . . . . . . . . . . . . . . . . 32.2 Shape Memory Alloy actuation . . . . . . . . . . . . . . . . . . . . . . 5

2.2.1 Shape Memory Effects . . . . . . . . . . . . . . . . . . . . . . 52.2.2 Actuation aspects of Titanium-Nickel alloys . . . . . . . . . . . 7

2.3 Electrostatic actuation . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3.1 Parallel-plate, comb-drive and curved-electrode actuators . . . 112.3.2 S-shaped film actuators . . . . . . . . . . . . . . . . . . . . . . 13

3 Heterogeneous integration 153.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1.1 Monolithic and hybrid integration . . . . . . . . . . . . . . . . 153.1.2 Heterogeneous integration . . . . . . . . . . . . . . . . . . . . . 17

3.2 Heterogeneous integration concepts and technologies . . . . . . . . . . 193.2.1 Transfer/direct and wafer-to-wafer/chip-to-wafer integration . . 193.2.2 Electrical interconnection . . . . . . . . . . . . . . . . . . . . . 213.2.3 Wafer-bonding techniques . . . . . . . . . . . . . . . . . . . . . 223.2.4 Releasing structures for actuation . . . . . . . . . . . . . . . . 24

4 Knife gate microvalves with bulk SMA microactuators 274.1 Integration of Titanium-Nickel shape memory alloy . . . . . . . . . . . 274.2 Gas microvalves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304.2.2 Knife gate microvalves . . . . . . . . . . . . . . . . . . . . . . . 34

4.3 TiNi sheet actuated knife gate valves . . . . . . . . . . . . . . . . . . . 354.3.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354.3.2 Fabrication - Integration of TiNi sheets . . . . . . . . . . . . . 354.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

viii Wafer-level heterogeneous integration of MEMS actuators

4.4 TiNi wire actuated knife gate valves . . . . . . . . . . . . . . . . . . . 384.4.1 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384.4.2 Fabrication - Integration of TiNi wires . . . . . . . . . . . . . . 394.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

5 Automated main distributing frames with S-shaped actuator switches 435.1 Switch units in automated main distributing frames . . . . . . . . . . 435.2 MEMS switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465.2.2 The S-shaped actuator switch . . . . . . . . . . . . . . . . . . . 47

5.3 The MEMS crossbar switch . . . . . . . . . . . . . . . . . . . . . . . . 485.3.1 The routing network . . . . . . . . . . . . . . . . . . . . . . . . 485.3.2 The crosspoint switches . . . . . . . . . . . . . . . . . . . . . . 505.3.3 The integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 505.3.4 Individual switch addressing . . . . . . . . . . . . . . . . . . . . 50

5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

6 Summaries of the Appended Papers 57

7 Conclusions 61

Acknowledgements 63

References 65

Paper reprints 79

LIST OF PAPERS ix

List of papersThe presented thesis is based on the following journal papers:

1. Out of plane knife gate microvalves for controlling large gas flowsS. Haasl, S. Braun, A. S. Ridgeway, S. Sadoon, W. van der Wijngaart andG. StemmeIEEE/ASME Journal of Microelectromechanical Systems, vol. 15, no. 5, pp. 1281–1288, Oct. 2006.

2. Wafer-scale manufacturing of bulk shape memory alloy microactuators based onadhesive bonding of Titanium-Nickel sheets to structured silicon wafersS. Braun, N. Sandström, G. Stemme and W. van der WijngaartIEEE/ASME Journal of Microelectromechanical Systems, accepted for publica-tion

3. Design and wafer-level fabrication of SMA wire microactuators on siliconD. Clausi, H. Gradin, S. Braun, J. Peirs, G. Stemme, D. Reynaerts andW. van der WijngaartIEEE/ASME Journal of Microelectromechanical Systems, accepted for publica-tion

4. Localized removal of the Au-Si eutectic bonding layer for the selective release ofmicrostructuresH. Gradin, S. Braun, G. Stemme and W. van der WijngaartIOP Journal of Micromechanics and Microengineering, vol. 19, no. 10, pp. 105014–105023, Oct. 2009.

5. Single-chip MEMS 5×5 and 20×20 double-pole single-throw switch arrays forautomating telecommunication networksS. Braun, J. Oberhammer and G. StemmeIOP Journal of Micromechanics and Microengineering, vol. 18, no. 1, pp. 015014–015025, Jan. 2008.

6. Row/Column addressing scheme for large electrostatic actuator MEMS switcharrays and optimization of the operational reliability by statistical analysisS. Braun, J. Oberhammer and G. StemmeIEEE/ASME Journal of Microelectromechanical Systems, vol. 17, no. 5, pp. 1104–1113, Oct. 2008.

The contribution of Stefan Braun to the different publications:

1 part of fabrication, all experiments, part of writing2 major part of design, major parts of fabrication and experiments, all writing3 part of design, fabrication, experiments and writing4 part of design, fabrication and experiments, major part of writing5 major part of design, all fabrication and experiments, major part of writing6 major part of design, all fabrication and experiments, major part of writing

x Wafer-level heterogeneous integration of MEMS actuators

The work has also been presented at the following international conferences :

1. Small footprint knife gate microvalves for large flow controlS. Braun, S. Haasl, S. Sadoon, A. S. Ridgeway, W. van der Wijngaart andG. StemmeThe 13th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems(TRANSDUCERS), Seoul, Korea, June 2005, pp. 329–332.

2. MEMS single chip microswitch array for re-configuration of telecommunicationnetworksS. Braun, J. Oberhammer and G. StemmeProceedings of the 36th European Microwave Conference (EUMW), Manchester,UK, Sep. 2006, pp. 811–814.

3. MEMS single-chip 5x5 and 20x20 double-switch arrays for telecommunicationnetworksS. Braun, J. Oberhammer and G. Stemme20th IEEE Int. Conf. on Micro Electro Mechanical Systems (MEMS), Kobe,Japan, Jan. 2007, pp. 811–814.

4. Smart individual switch addressing of 5x5 and 20x20 MEMS double-switch ar-raysS. Braun, J. Oberhammer and G. StemmeThe 14th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems(TRANSDUCERS), Lyon France, June 2007, pp. 153–156.

5. Robust trimorph bulk SMA microactuators for batch manufacturing and inte-grationS. Braun, T. Grund, S. Ingvarsdottír, W. van der Wijngaart, M. Kohl andG. StemmeThe 14th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems(TRANSDUCERS), Lyon, France, June 2007, pp.2191–2194.

6. Wafer-scale manufacturing of robust trimorph bulk SMA microactuatorsN. Sandström, S. Braun, T. Grund, G. Stemme, M. Kohl and W. van der Wi-jngaartProceedings of the 11th Int. Conf. on new Actuators (ACTUATOR), Bremen,Germany, June 2008, pp. 382–385.

7. Microactuation utilizing wafer-level integrated SMA wiresD. Clausi, H. Gradin, S. Braun, J. Peirs, G. Stemme, D. Reynaerts andW. van der Wijngaart22nd IEEE Int. Conf. on Micro Electro Mechanical Systems (MEMS), Sorrento,Italy, Jan. 2009, pp. 1067–1070.

8. Selective electrochemical release etching of eutectically bonded microstructuresH. Gradin, S. Braun, M. Sterner, G. Stemme and W. van der WijngaartThe 15th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems(TRANSDUCERS), Denver, USA, June 2009, pp. 743–746

LIST OF PAPERS xi

9. Full wafer integration of shape memory microactuators using adhesive bondingN. Sandström, S. Braun, G. Stemme and W. van der WijngaartThe 15th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems(TRANSDUCERS), Denver, USA, June 2009, pp. 845–848

xii Wafer-level heterogeneous integration of MEMS actuators

1 INTRODUCTION AND STRUCTURE 1

1 Introduction and structureThis thesis presents research in the field of microelectromechanical systems (MEMS),also referred to as micromachines in Japan or Micro Systems Technology (MST) inEurope. MEMS technology uses the tools and techniques that were developed for theIntegrated Circuit (IC) industry and allows for high volume parallel production ofdevices, potentially resulting in low fabrication costs per device. MEMS technologyincludes components with typical sizes between 1 to 100 μm (1 μm = 0.001 mm),which are combined to form MEMS devices such as pressure sensors, inertial sensors,switches, pumps, valves and many more with dimensions in the mm range.

While IC devices can be considered as the ’brain’ of a microsystem, MEMS de-vices provide the ’arms’ and the ’eyes’ to allow the IC device to sense and manipulatethe environment. This thesis focuses on the ’arms’, i.e. actuators which typicallyconvert electrical energy into mechanical movement. Methods were developed to in-tegrate actuators with silicon structures to fabricate microvalves for controlling largegas flows and crossbar switches for automating parts of copper-wire telecommunica-tion networks. The actuators are not integrated using the conventional monolithicfabrication, but using concepts of heterogeneous integration. Thus, the actuators arefabricated separately and finally bonded onto the silicon structures to functionalize.

This thesis is divided into two parts. The first part provides detailed backgroundinformation and informative references to facilitate a better understanding of thesecond part, which contains the appended journal applications.

The first part contains four chapters.In chapter 2, common MEMS actuator technologies are introduced with focus on

electrostatic and shape memory actuation, which are crucial elements of the MEMSdevices presented later on.

Chapter 3 introduces heterogeneous integration with reference to the other inte-gration methods, which are monolithic and hybrid integration. The chapter providesbackground information about heterogeneous integration concepts and technologiesincluding wafer-bonding methods and releasing of structures to be manipulated byintegrated actuators.

Chapter 4 introduces the concept of knife gate gas microvalves, methods to het-erogeneously integrate bulk shape memory sheets and wires for actuation and thecombination of both. Other microvalve types and shape memory integration methodsare discussed. First prototypes are presented and their performance is compared toother microvalves with respect to performance per consumed silicon footprint area.The work is discussed and an outlook on future work is given.

Chapter 5 introduces a MEMS crossbar switch for automating parts of copper-wiretelecommunication networks. The chapter presents in detail the application, the inte-gration of arrays of electrostatically actuated switches and a method to individuallyactuate one out of the 400 switches in the array without underlying CMOS addressingcircuits. The work is discussed and an outlook on future work is given.

2 Wafer-level heterogeneous integration of MEMS actuators

2 MEMS ACTUATORS 3

2 MEMS actuatorsActuators and sensors are transducers which convert one type of energy into anotherone. Sensors are mostly used to measure a physical parameter and report it in formof an electrical signal. Actuators work the other way, typically converting electricalenergy into mechanical work output. The term ’MEMS actuators’ summarizes allactuators which typically are submillimeter sized and fabricated with MEMS tech-nologies.

The following chapters will provide a short summary of the main MEMS actu-ation principles, with focus on electrostatic and shape memory actuation which areimportant parts of this thesis.

2.1 Common microactuation mechanismsMEMS technology offers a wide range of microactuators which can be classifiedinto electrostatic, thermal, piezoelectric, magnetic and shape memory alloy actua-tion methods. Table 1 and the following paragraphs present a brief overview of theprinciples according to [1]. A more detailed review of all the different microactuatorsand their principles would go beyond the scope of this thesis and the interested readeris referred to literature for deeper information [1, 2, 3, 4, 5, 6, 7].

The piezoelectric actuation utilizes the coupling of mechanical deformation andelectric polarization in certain materials. When applying a mechanical stress to thematerial, it generates an electrical voltage. This effect is called direct piezoelectriceffect and is used for sensing and energy harvesting applications. However, for actua-tion the inverse piezoeffect is deployed: by applying a voltage the material generatesmechanical movements. The stroke of piezoelectric actuators is in general very small,however, relatively large forces can be obtained with a very precise displacement res-olution. Furthermore, these actuators are fast, allowing for high cycling frequenciesand making them very feasible for repetitive actuation. An example of a commercialapplication deploying the advantages of piezoelectric actuation is the smallest PiezoLEGS® linear motor, which allows for linear travel distances limited only by thelength of the displaced element (comes with a 50 mm long element) at a speed of20 mm

s with 10 N force and a resolution smaller than 1 nm.For MEMS applications, the integration of piezoelectric actuators is challenging.

After the deposition the ceramic material, which most commonly is lead zirconatetitanate (PZT), must be sintered at high temperatures (> 600 °C), which limits the


Table 1. The different principles commonly used for microactuation. The real work densitymight be substantially lower.

Method Principle Work density

Electrostatic Attractive force between bodies withdifferent electrostatic charges.

≈ 105 Jm3

Piezoelectric Shape change under an electric field (inversepiezoeffect).

≈ 1.2·105 Jm3

Thermal Thermal expansion of single material(includes sealed liquid) or difference in CTEbetween two materials or phase change.

≈ 5·106 Jm3

Magnetic Interaction with magnetic fields. ≈ 4·105 Jm3

Shape memoryalloy

Temperature dependent crystal phasetransformation with macroscopic shapechange. Belongs technically to thermalactuation.

≈ 107 Jm3

range of compatible materials and processes. Furthermore, the ceramic shrinks duringsintering, resulting in very high mechanical stresses if the ceramic is constrained bythe substrate prior to sintering.

Magnetic actuation is based on interaction with magnetic fields which are gener-ated by either permanent magnets or coils. The energy density of magnetic actuationis in the same order of magnitude as electrostatic actuation. For some applicationsmagnetic actuation is considered as an alternative to electrostatic actuation becauseof advantages such as non-existence of electric discharges, possibility of operation inliquid and long-range forces. However, integrating magnetic field sources or low reluc-tance materials on-chip is a major challenge; the coils are three-dimensional structureswhich are complicated to fabricate and the hard or soft magnetic materials are diffi-cult to integrate with MEMS structures [8].

Thermal actuation is based on volume or phase change of materials upon heatingor cooling. The displacement is analog to temperature change and thereby sensitiveto environmental influences. Typically, thermal actuators provide large stroke andlarge forces, however, with a rather limited displacement resolution. Heating of theactuators can be very fast by a high applied power, however, cooling is in most casespassive and the cooling time severely slows down the actuator and decreases theactuation frequency.

Examples of MEMS thermal actuation approaches are the ’heatuator’ [9] usingone material, bimorph using different materials [10], thermopneumatic actuation andshape memory alloys. The ’heatuator’ is a U-shaped lateral thermal actuator fabri-cated in one material, with one arm of the ’U’ considerably narrower than the other.Current is passed through the actuator, and the higher current density in the narrower

2 MEMS ACTUATORS 5

’hot’ arm causes it to heat and expand more than the wider ’cold’ arm. The armsare joined at the free end, which constrains the actuator tip to move laterally in anarcing motion towards the cold arm side. The bimorph actuation scheme is based onthe difference in coefficients of thermal expansion (CTE) between two joined layers ofdifferent materials. When changing the ambient temperature, one layer expands orshrinks more than the other and the resulting interface stress causes bending of thestack. The amount of bending can be controlled by choosing the appropriate CTEcombination and by the applied temperature/electrical power. The thermopneumaticactuation scheme is based on the thermal expansion of a fluid sealed inside a cavity.Heating causes volume expansion or even phase change, exerting a large force on thecavity walls and causing a bending of a deformable membrane. Shape memory alloysare materials which undergo a phase change upon temperature changes. This phasechange comes with a macroscopic shape change of the device and provides the highestenergy density of all MEMS actuators. SMA actuation is an important part of thisthesis and is presented in detail in the next subsection.

2.2 Shape Memory Alloy actuationTechnically, shape memory alloy (SMA) actuators are thermal actuators, since ther-mal energy triggers a crystal phase change in the material. However, since SMAactuation is an important part of this thesis, it is presented on its own.

2.2.1 Shape Memory Effects

Structures made from Shape Memory Alloys exhibit the shape memory effect (SME),which is the ability of certain materials to ’remember’ their initial shape after theyhave been deformed. Take a spring made out of SMA and pull it. It will be easy todeform and it will stay deformed. Now heat the spring above a specific transformationtemperature and it will rapidly recover the initial shape.

The underlying mechanism of the shape memory phenomenon is a martensite toaustenite and vice versa phase transformation. Figure 1 illustrates the different crystalstates and their connected macroscopic shapes by means of a SMA spring. In the hotstate, the SMA crystal structure thermodynamically prefers a more ordered phase andtransforms to the austenite crystal form with the macroscopic shape connected to it.This phase is also called parent phase and it is possible to set the desired macroscopicshape by a specific treatment involving mechanical constraining and heating. Whenthe spring is cooled again and there is no external force applied, it will still havethe same shape as in the hot state, however, its crystal is not in the cubic formanymore. Instead, the layers in the material are tilted, with the tilting directionalternating between each layer. Because of these alternated tilts, the spring remainsits shape even though the crystal form has changed. The material of the spring isnow in a macroscopically non-deformed, low-temperature phase referred to as self-accommodated martensite [5] or ’twinned’ martensite [11], since its characteristicalternated layers are called ’twins’. In this phase, the material features a very low yieldstrength and can easily be plastically deformed after straining it above a very narrowelastic strain range - therefore the spring can be pulled very easily. After the externalforce is removed, the spring will stay deformed except for a very small elastic strain


austenite phaseparent phase

crystal phase macroscopic shape

mechanical load

coolingheating

two-way SME

cooli

ng

heati

ng

self-accomodated martensitetwinned martensite


detwinned martensite


unload

super-elasticity

only for T > Af

load

heating

one-way SME

Figure 1. Illustration of the transformation processes in martensitic transformation, in-cluding all the shape memory effects.

recovery. During the deformation, the alternated layers of the twinned martensite aremoved and the macroscopic shape of the structure is changed. Now the material isstill in a martensitic phase and easily deformable. Since the alternated layers calledtwins are removed, this martensitic phase is called ’detwinned’ martensite.

Figure 2 displays the hysteresis behavior of the phase transformation versus thetemperature. The transformation from martensite (cold) to austenite (hot) starts atthe austenite start temperature As and finishes at the austenite finish temperatureAf . Vice versa, the transformation from austenite (hot) to martensite (cold) starts atthe martensite start temperature Ms and finishes at the martensite finish temperatureMf . Hence, there is no single transformation temperature. However, in practice it isreferred to a temperature T0, which is the thermodynamic equilibrium temperatureof martensite and austenite state (T0= MS+As

2 ) [5].

There are three different shape memory effects, which are illustrated in the overviewin figure 1. In the example above, the spring must be deformed with an external forceand performs work only in one direction from detwinned martensite to austenite.Thus, this effect is called the one-way effect. However, the material can be trained toassume a certain shape in the cold state [12,13]. Then, the crystal transforms directlybetween austenite and detwinned martensite. Hence, the material performs work intwo directions and this effect is called the two-way effect. The third effect is calledsuperelasticity or pseudoelasticity and is only present if the SMA is always at tem-peratures above the austenite finish temperature Af . Then, if an externally appliedstress overcomes a critical stress, the crystal transforms from austenite to detwinned

2 MEMS ACTUATORS 7

austenite

martensite

crystal phase

temperature T

Mf Ms T0 AfAs

Figure 2. Illustration of the hysteresis behavior of the martensitic transformation with theassociated temperatures: austenite start As, austenite finish Af , martensite start Ms andmartensite finish Mf . The temperature T0 is the thermodynamic equilibrium temperatureof martensite and austenite state.

martensite. When the stress is removed, the crystal immediately transforms back toaustenite. In this case, the spring in the example above will immediately, and withoutextra energy supply, recover the straight shape when the deforming external force isremoved.

2.2.2 Actuation aspects of Titanium-Nickel alloys

The shape memory phenomenon was first discovered in the 1930s in brass alloys. In1962, Buehler and his colleagues found the shape memory effect in alloys of Nickeland Titanium [14] and in honor of their employer they named these alloys NiTiNOL(Nickel Titanium Naval Ordinance Laboratory). Nowadays, these allows are alsoknown under the acronyms TiNi (Titanium-Nickel) or NiTi (Nickel-Titanium). Inthis thesis, the term TiNi is used. Besides TiNi, there are a number of other mate-rials showing the shape memory effect, such as other metallic alloys, polymers andeven bacteria [15, 16]. However, TiNi based SMA devices are dominating the marketbecause of several advantages over other alloy systems. TiNi alloys allow to adjustthe transformation temperature T0 over a wide range only by changing the ratio ofnickel atoms. If the alloy consists of half nickel and half titanium atoms, the trans-formation occurs near 100 °C. However, adding slightly more nickel atoms decreasesthe transformation temperature to below 0°C. Furthermore, these alloys can be fab-ricated with standard metalworking techniques, they exhibit better shape memorystrain performance than other known alloys and consist of the affordable elementsNickel and Titanium. TiNi is a biocompatible material, making it interesting espe-cially for medical applications. All the following reflections and work presented inthis thesis are based on TiNi alloy.

SMA actuation is generally based on the one-way effect, i.e. when heated upondeformation the structure recovers its initial shape, yet upon cooling the shape doesnot change by itself. The approaches to utilize the one-way SME are usually summa-rized in three different categories, depending on the load which is applied upon theSMA during shape recovery [5, 11, 17].

1. In free recovery, a deformed SMA device is not constrained by any external loadduring the shape recovery and therefore the SMA does not provide any force.


2. In constrained recovery, the shape recovery of a deformed SMA device is blockedby an external constraint, triggering large forces from the SMA.

3. In cyclic work production, the shape recovery is constrained. However, uponheating the SMA can overcome the external force for the shape recovery. Uponcooling, the external force deforms the SMA again until the next temperaturecycle. The external force is called bias spring or cold-state reset.

The combination of SMA and bias spring described under cyclic work production isthe basis for SMA microactuators. The methods providing the cold-state reset canbe summarized as intrinsic and extrinsic methods [18].

Using intrinsic methods, the crystal of the material is modified to prefer a certaincold-state crystal orientation, which results in a preferred shape of the structure in thecold-state. An example of an intrinsic cold-state reset is the two-way shape memoryeffect [19, 20], where a cold-state shape is trained into the material using long-termcycling processes. However, compared to extrinsic cold-state reset methods, the two-way effect is very unstable [19,20], exhibits considerably less recoverable deformationand furthermore the required training process is difficult to integrate with a batchfabrication process for MEMS applications [5, 21]. Therefore this method will not befurther discussed in this thesis.

Most of the SMA actuators deploy extrinsic biasing methods, where the SMAmaterial is coupled with an additional mechanical element. A widely used biasingscheme, especially in MEMS applications, is coupling of the SMA with an externalbiasing spring element. This topic will be addressed in detail below. Another in-teresting scheme is the antagonistic biasing, where two SMA elements are coupledtogether. While element A remains cold and very easy to deform, element B is heatedand pulls the cold element A without requiring high forces. Then, after cooling, bothelements maintain their current shape until element A is heated, thereby deformingthe cold and soft element B. Since the SMA bias spring is very easy to deform overrelatively large strains (of course only within the elastic range), large deflections canbe obtained. Yet, in MEMS applications it is difficult to couple two SMA elementsin combination with a good thermal isolation between them.

The achievable work density of the TiNi is very much depending on the load case.Table 2 [22] shows the different cases. Under pure tension or compression load, thehighest forces can be obtained, however, with relatively small displacements. Largerdisplacements, yet with lower forces, can be obtained under torsion or bending load.Bending load provides the lowest energy density of the three different load cases, sinceonly fractions of the material are used for work production.

Another important aspect to consider when designing the actuator is the fatigueof the material. Fatigue in TiNi is usually divided into structural and functionalfatigue [23]. Structural fatigue refers to the mechanical failure of the TiNi aftercyclic loads, similar to any other engineering material. But unlike normal engineeringmaterials, shape memory alloys show different properties in different temperatureranges, which also influences the fatigue characteristics.

Functional fatigue refers to a decrease in functional properties, which is the shape

2 MEMS ACTUATORS 9

Table 2. Comparison of work density and energy efficiency of TiNi wires for three differentload cases [22].

Load case Work density ( Jkg ) Energy efficiency (%)

Tension/Compression 466 1.3Torsion 82 0.23Bending 4.6 0.013

Table 3. Allowable stress and strain for a targeted amount of actuation cycles [22].

Cycles Max.strain (%) Max. stress (MPa)

1 8 500102 4 275104 2 140

> 105 1 70

recovery of the TiNi during cyclic loading. The functional fatigue is of great interest,since it defines how many cycles the actuator can be operated depending on the stressand strain applied to the material. Table 3 [22] shows some experimentally evaluatedbenchmark numbers for TiNi wires. When straining the wire with the maximumpossible 8% or stressing it with the maximum possible 500 MPa, only one shaperecovery cycle can be obtained. To maximize the number of actuation cycles, theapplied strain should be below 1% or the applied stress should be below 70 MPa.

Thermal energy must be provided to trigger the shape recovery of the SMA. Anoption is to vary the ambient temperature. However, for cyclic actuation purposesthis is rather impractical. The TiNi can be heated by electrically contacting and Jouleheating the material itself. However, the stable oxide on the TiNi makes electricalcontacting complicated. Therefore, especially for MEMS applications, the heating issometimes performed indirectly using a separate resistive heater, which can be con-tacted in a simpler way. The voltages needed to operate SMA microactuators arecompatible with microelectronics, however, high currents are necessary to provide therelatively high power for heating of the SMA.

During operation, the TiNi transforms between two states and displays hystere-sis behavior as described earlier. Because of the two stable states, the SMA is verysuitable for applications that require digital mode operation of the actuator. Forsuch applications, the hysteresis behavior is potentially of advantage; the thermalenergy necessary to maintain the austenite state is lower than the initial austenitestart temperature As, which defines the stable state very well even for an unstablethermal energy supply. For applications requiring precise analog-like control over thedisplacement of the actuator, there are several controlling solutions. One method isthe model-based loop, which is based on extensive modeling of the materials behaviorto reduce or compensate the hysteresis effect. However, the necessary material param-


d

d0 d02

3

Vpull-out Vpull-inV

gradual deflection

pull-in

pull-out

maintains currentstate

pulledout

pulledin

(c)(a)

d0

moveable plate, Area A

fixed plate

(b)

d VFel

Fs

k

Figure 3. Diagram illustrating the operational behavior of an electrostatic actuator: (a)and (b) Parallel-plate capacitor model showing the principle of electrostatic actuation withand without applied voltage, respectively; (c) Diagram illustrating the hysteresis behaviorof an electrostatic actuator, showing the typical pull-in and pull-out characteristic.

eters must be experimentally evaluated. Another method, the feedback-loop, relies onsensing of either position [24], temperature [25] or electrical resistance [26] of the SMAstructure as input. Both position and temperature sensing require additional devicesand especially the temperature sensing is impractical due to temperature disturbancesin an open environment. The control based on electrical resistance sensing is very in-teresting since it utilizes the smart material capabilities of the SMA. Yet anotherinteresting approach is to keep the digital mode operation with all its advantages, butsegmenting the SMA element to quantify the deflection [27].

This overview is far from being complete, therefore the interested reader is referredto [28] for more information on this topic.

2.3 Electrostatic actuation

The electrostatic actuation principle relies on the attraction force between bodies hav-ing different electrostatic potential caused by a charge inbalance. A simple exampleof an electrostatic actuator is a parallel-plate capacitor, as illustrated in Figure 3a,with one fixed plate and the other plate suspended by a mechanical spring with aspring constant k at an initial distance d0. Applying a voltage V between the twoplates results in a vertically attractive electrostatic force, which pulls the moveableplate towards the fixed plate (Figure 3b). Using this simplified model and neglectingfringe-fields, the electrostatic force Fel between the plates can be calculated as [29]

Fel =12ε0εr

A

d2V 2 (1)

with ε0 the permittivity of free space, εr the effective relative permeability, A theoverlap area of the two plates and d the distance between the two plates.

This formula is the basic formula for all electrostatic actuators and shows that theelectrostatic force grows quadratically with decreasing distance between the plates,which makes electrostatic actuation very interesting for MEMS applications with verysmall gap distances below tens of micrometers.

2 MEMS ACTUATORS 11

2.3.1 Parallel-plate, comb-drive and curved-electrode actuators

For electrostatic actuators based on the parallel-plate concept there are several issuesto consider. The electrostatic force Fel is counteracted by the mechanical spring forceFs, which is calculated as:

Fs = −k(d0 − d) (2)

For practical designs with a low actuation voltage, small electrode area and asufficiently stiff mechanical spring, the initial plate distance d0 must be small, whichresults in small travel distances d (strokes) of typically a few micrometers for themoveable plate. Furthermore, the range in which the stroke of the moveable platecan be controlled is limited. Figure 3c illustrates the deflection of the moveable plateduring a full operation cycle. With increasing actuation voltage, the gap between theplates gradually decreases and the two forces Fel and Fs will settle in an equilibrium.However, with decreasing d, the electrostatic force grows quadratically, whereas thecounteracting spring force only grows linearly. At distances smaller than the criticaldistance d = 2

3d0, there no longer exists an equilibrium between the forces and themoveable plate snaps down to the fixed plate. To avoid an electrical short-circuit aftersnap-down, there must be an electrical isolation layer or at least ’dimples’ (distanceholders) between the plates. The critical distance is independent of the geometricalparameters of the actuator and the voltage at which the plate snaps down is calledthe pull-in voltage, or Vpull−in (figure 3c). After the pull-in, the gap d is drasticallyminimized and therefore, when reducing the applied voltage after pull-in, the electro-static force remains larger until a force equilibrium is reached again. When furtherreducing the applied voltage, Fs overcomes Fel and the moveable plate is pulled out.Accordingly, the voltage at which the pull-out occurs is called Vpull−out (figure 3c).

Some applications require an analog behavior of the actuator and there are effortsto extend the limited analog controllable stroke of parallel-plate actuators [30,31,32].However, there are also many applications demanding a digital mode operation ofthe actuator, such as electrical and optical switches which alternate between theON and the OFF state. In these applications, the hysteresis behavior is actually ofadvantage; the voltage necessary to maintain the pull-in state is lower than the initialactuation/pull-in voltage, which defines the switching state very well even at unstablecontrol voltages.

In contrast to the parallel-plate actuator, the ideal comb-drive actuator showsno pull-in and hysteresis behavior since the plates are not moving perpendicularly,but parallel to each other and thereby keep the plate distance d constant during theoperation. A second fixed plate is added and the moveable plate is interdigitatedbetween the two fixed plates with an initial lateral overlap x0. Figure 4 illustrates theactuator, which is called ’comb-drive’ since the interdigitated finger-like structureslook like the teeth on a comb.

Applying an actuation voltage results in several forces, as illustrated in figure 4b.The two vertical force components Fel,y keep the moveable finger centered betweenthe stationary fingers and the lateral force component Fel,x pulls the moveable fin-ger towards the fixed fingers and counteracts the lateral bias spring with the spring


moveable plate, thickness t

d

fixed plate, thickness t

x0

fixed plate, thickness t

d

Fel,x

Fel,y

Fel,y

Fs

x kV

(a) (b)

Figure 4. Diagram illustrating the operational behavior of an electrostatic comb-driveactuator, without (a) and with (b) applied actuation voltage.

constant k.The lateral force Fel,x is independent of the plate overlap and remains constant

with increasing plate overlap. The distance between the capacitor plates also remainsconstant, which allows for large analog controllable stroke, only limited by the elasticrange of the bias spring.

However, the large stroke comes at a cost. The fingers are usually fabricated byvertical etching into the silicon substrate using deep reactive ion etching (DRIE). Asfor all electrostatic actuators, the distance d between the fingers should be as small aspossible and the electrode area A as big as possible. Consequently, high aspect ratioprocesses are necessary to produce structures with a minimal distance in between.Since the aspect ratio and the resulting initial distance is limited, the only way toincrease the electrode area and the force is the massive parallelization of comb struc-tures, at the cost of silicon footprint.

Both large stroke and large force are provided by curved-electrode actuators, whichutilize a flexible beam opposite a fixed electrode. Figure 5 illustrates the principle.One end of the flexible beam is clamped with a very short distance to the fixedelectrode. One of the two electrodes is a curved electrode, shaped in a way thatthe electrode distance is gradually increasing from the clamped end to the free end.Upon applying an actuation voltage, the narrow gap at the clamped end results inlarge forces and the flexible beam is pulled in. As for the parallel-plate actuator,electrical isolation between the beams is necessary to avoid an electrical shortcircuit.The point of pull-in is moving along the fixed electrode in a zipper-like way andtherefore these actuators are also referred to as ’zipper-actuators’. Another nameis ’touch-mode’ actuator, since these actuators utilize the pull-in and the plates aretouching each other only separated by a thin electrical isolation layer or stoppers.The combination of small plate distance at the clamped end, the touching mode withvery thin gaps between the electrodes and the large distance at the free end resultsin large forces and a large stroke, making this actuation scheme very interesting forMEMS applications.

For the most common zipper actuators, the moveable part is moving either lat-erally or vertically, as illustrated in figure 5 [33]. Besides the orientation of theactuation, the two configurations also differ in the arrangement of fixed and moveableplate as well as in their fabrication. In the lateral zipper approach, the fixed electrode

2 MEMS ACTUATORS 13

(a) (b)

d0

silicon

d0

moveable, curvedelectrode

fixed, straightelectrode

moveable, straight

electrode

fixed, curved

electrode

Figure 5. The two most common fashions of zipper actuators: (a) lateral zipper and(b) vertical zipper. The figure is modified from [33].

is curved and the moveable electrode is a straight cantilever. The fabrication is fairlysimple with one photolithographical mask, vertical etching into the device layer ofa SOI wafer and sacrificially underetching the buried oxide to release the moveablebeam. However, as for the comb-drive actuator, the initial gap and the electrodearea are limited by the aspect ratio of the fabrication process. In the vertical zip-per approach (illustrated in figure 5b and 6a), the fixed electrode is straight and themoveable electrode is curved, typically fabricated using surface micromachining. Thecurvature of the moveable electrode results of a controlled, fabrication process relatedstress gradient and because of the curvature, these kind of actuators are also called’curled actuators’. In contrast to the lateral approach, the initial gap at the clampedend can be very narrow by utilizing a thin sacrificial layer and the electrode area canbe very large, resulting in large forces at relatively low actuation voltages. However,the stress gradient in the moveable electrode and the resulting spring tension counter-acting the electrostatic force is difficult to control. Furthermore, stiction could occurbetween the large area electrodes in close contact.

2.3.2 S-shaped film actuators

The stress gradient in the bending electrodes of standard vertical zipper actuatorsis difficult to control. A large stress gradient results in a spring with a high pre-tensioning, counteracting the electrostatic force and resulting in larger actuation volt-ages to pull in the electrode. A thin and soft membrane would decrease the necessaryactuation voltages, however, a weak spring cannot provide a reliable pull-out andsuspension of the electrode.

A solution allowing for a thinner and softer membrane is the incorporation of asecond fixed electrode at the free end of the membrane, providing a second zipperactuator, as illustrated in figure 6b. The resulting double-zipper actuator providesactive actuation of the film in both directions, allowing for a very flexible membranewith a low stress gradient and thereby potentially reducing the actuation voltage yetstill allowing for a large stroke. A MEMS concept of such an actuator is illustratedin figure 6b. First, a single zipper is fabricated with a thin and flexible membrane,


touch-mode, pull-in ’zips’ along fixed electrode

V Vfixed electrode

d0

pre-stressed flexible electrode

fixed electrode

2nd fixed electrodetwo ’zippers’, d0 defined by thin electrical

isolation layer between the electrodes

V

d0

d0

(a) single ’zipper’ actuator

(b) double ’zipper’ or S-shaped film actuator

Figure 6. Illustration of (a) single vertical zipper actuator and (b) double vertical zipperor S-shaped film actuator.

yet still with sufficient bending of the free end. Then, the second electrode is addedfrom the top and pushes the membrane in contact with both electrodes and creatingthe characteristic S-shape of the membrane which inspired the name of the S-shapedactuator.

To allow the membrane to move up and down, the two fixed electrodes must bekept at a distance to each other with an intermediate spacer. The thickness of thisspacer allows to tune the distance the membrane can move up and down between thetwo electrodes, which defines the stroke of the actuator.

In summary, this concept comes with a set of advantages. The touch-mode actu-ation, with very small initial gaps in both directions, in combination with a thin andflexible membrane potentially results in very low actuation voltages. The stroke ofthe actuator is basically only limited by the tuneable spacing between the two fixedelectrodes.

The S-shaped actuator was shown in 1997 for a gas valve with dimensions in themillimeter range [34]. Another work [35, 36, 37] utilized the assembly concept of theS-shaped actuators to fabricate an RF MEMS switch by fabricating a single zipper ac-tuator with metal contacts on one substrate and combining it with a second substrate,which contained the second fixed electrode and signal lines to be interconnected. Inthis work, the spacing between the substrates was provided by a polymer ring, whichalso encapsulated and packaged the switch.

3 HETEROGENEOUS INTEGRATION 15

3 Heterogeneous integration

Heterogeneous integration evolved from monolithic and hybrid integration and refersto the wafer-level integration of different materials, technologies or devices. In mono-lithic integration a device is fabricated in one piece while in hybrid integration a deviceis fabricated by interconnecting several separate pieces. The following sections intro-duce the different integration methods, followed by technical background includingmethods for wafer-to-wafer bonding, vertical electrical interconnection and releasingof structures for actuation.

3.1 Introduction

The following sections introduce the concepts of monolithic, hybrid and heterogeneousintegration. Heterogeneous integration is of high interest for the integration of MEMSand IC and therefore the integration technologies are introduced by means of thespecific example of integrating MEMS materials onto IC circuits.

3.1.1 Monolithic and hybrid integration

In monolithic integration, devices are fabricated from one substrate (monolithic =made from one piece). All the processing is typically performed on wafer-level andafter the fabrication the wafer is diced into discrete devices (figure 7a), which areready for further application.

As an example, MEMS and IC are monolithically integrated by combining andcustomizing the MEMS and IC manufacturing processes. The main technical ad-vantage of monolithic integration of MEMS and IC is the high integration density;electrical interconnections between MEMS and IC are very short, reducing electricalnoise and allowing for the handling of small signals. However, monolithic integrationof MEMS and IC is relatively complicated [38,39,40,41,42], since MEMS technologycan require IC incompatible material deposition processes and/or temperatures above450 °C, which is not allowed for the IC components.

A solution to avoid these problems is the hybrid integration (hybrid = combinationof different parts), where the devices are fabricated on separate substrates, which arethen diced into single chips and combined with each other on chip level (figure 7b).

As an example, MEMS and IC are hybrid integrated by fabricating on separatesubstrates, which are then diced into single MEMS and IC chips. Conventionally,


(b) hybrid integration

(a) monolithic integration

(c) heterogeneous integration

IC 1 IC 2 IC 3MEMS 1 MEMS 2 MEMS 3

wirebond

adhesive

Device 1 Device 2 Device 3carrier substrate carrier substrate carrier substrate

Device 1 Device 2 Device 3

bonding

MEMS+IC

dicing

dicing dicing

pick-and-place assembly

MEMS

MEMS

IC

IC

dicing

removing source substrate

substrate

substrate substrate

target substrate

source substrate

Device 1 Device 2 Device 3

Figure 7. Simplified schematic illustrations of the different methods for integrating MEMSwith IC: (a) monolithic integration, (b) hybrid integration and (c) heterogeneous integration.


(a) (b) (c) (d)

MEMS ASIC

wirebonds

Figure 8. Example of wire bonding based hybrid integration of a MEMS inertial sensorwith an ASIC for automotive applications [49]. (a Leadframe. (b) The MEMS sensor (top)and the ASIC (bottom) are adhesively mounted onto the leadframe. Using wirebonding, theMEMS is electrically connected to the ASIC and the ASIC is connected to the leadframe.(c) Packaged by plastic molding. (d) The leadframe pins are punched free and shaped,resulting in a chip ready for integration in a larger system.

these chips are glued beside each other on a carrier substrate and electrically inter-connected by wire-bonding (see example in figure 8) or by connections integrated inthe substrate (Multi Chip Modules [43]). Alternatively, they are stacked on top ofeach other [43,44] using through-substrate-vias (TSV) [45,46,47,48] for vertical elec-trical interconnection. The main technical advantage is the uncomplicated integrationof different technologies, materials or devices. However, there are applications wherehybrid integration is not feasible due to cost-efficiency reasons, limited integrationdensity and parasitic signal noise from the long electrical interconnections.

Heterogeneous integration allows to combine the two approaches and is presentedin the next section.

3.1.2 Heterogeneous integration

Typically, heterogeneous integration is utilized for integrating materials or technolo-gies which otherwise are very difficult to combine or even incompatible with eachother. However, heterogeneous integration also allows to divide the fabrication ofMEMS devices into several separate sub-structures, which are optimized for a certainaspect and finally combined to one device. Both aspects are included in this thesis.Chapter 4 addresses the integration of an incompatible actuator material with siliconstructures and chapter 5 addresses the separate fabrication of MEMS actuators arraysand their integration to functionalize another MEMS device.

Heterogeneous integration follows the same basic concept as hybrid integration.The devices to be integrated, or parts of them, are fabricated separately. However, incontrast to hybrid integration, the two different substrates are integrated on wafer-level by bonding them on top of each other, followed by removing the substrate ofthe integrated device and dicing into single chips (figure 7c). This wafer-level hy-brid integration method allows to combine the advantages of hybrid and monolithicintegration such as separate fabrication and wafer-level processing including high in-tegration density and short electrical interconnections between the devices.

An example to demonstrate all the benefits of heterogeneous integration is the re-placement of the mirror material of micromirror arrays from aluminum to monocrys-


’dummy IC’ substrateIC substratefabricate actuation electrodes

spin-on and pattern sacrificial layer

deposit mirror material by sputtering aluminum

pattern mirror material to form micromirrors

remove sacrifial layer

(a) monolithic (b) heterogeneous

fabricate actuation electrodes

spin-on and pattern sacrificial layer

deposit mirror material by bonding SOI – wafer and removing handle wafer and buried oxide

remove sacrifial layer

SOI - wafer

pattern mirror material to form micromirrors

electroplate vias for electrical inter-connection and mechanical clamping

Figure 9. Simplified illustrations of the fabrication of micromirrors: (a) monolithicallyfabricated with aluminum mirrors [50] and (b) heterogeneous integration of a silicon layerfor the mirrors [51].

talline silicon. The famous digital micromirror device (DMD) for projectors fromTexas Instruments [52] features an array of up to 2048×1152 micromirrors and eachof these mirrors must individually addressable, which for the DMD is performed usinga dedicated IC circuit. Using hybrid integration of the mirrors and the IC would notbe feasible since more than one million interconnection wires would be necessary tocontrol each mirror. Therefore, the mirrors are monolithically integrated on top ofthe IC by sputtering and patterning aluminum as mirror material. Figure 9a [50]shows a simplified example of such a process. Each mirrors control electrode is ad-dressed by a memory cell directly underneath, which eliminates the need to routeindividual control wires underneath the array. However, after repeated or prolongedmirror actuation, the aluminum mirrors display hysteresis and memory effects, whichcan be problematic for applications requiring analog mirror deflections. These issuescan be eliminated by utilizing monocrystalline silicon for the mirrors. Furthermore,the achievable optical quality, the surface roughness and the uniformity of monocrys-talline silicon surfaces is superior compared to most other surfaces. However, since theprocess temperatures for the material deposition onto IC is limited to about 450 °Cto avoid damage to the electronic circuits, many high-performance MEMS materialssuch as monocrystalline semiconductors cannot be monolithically integrated.

A solution is the heterogeneous integration of monocrystalline silicon mirrors ontothe IC driving circuitry by transferring the thin silicon device layer of a SOI-waferusing a IC compatible transfer bonding process. The involved temperatures are alwaysbelow 450 °C. Figure 9b shows a simplified example of such a process [53,54,55,56,51]


and the analogy to the monolithic fabrication of micromirror arrays.Heterogeneous integration was enabled with the advent of wafer bonding tech-

nologies and is an important technology for the ’More than Moore’ trend to integratenon-IC functions onto IC devices in order to increase their capabilities beyond MooresLaw [57]. More information about the technology behind heterogeneous integrationis presented in the following section.

3.2 Heterogeneous integration concepts and technologies

3.2.1 Transfer/direct and wafer-to-wafer/chip-to-wafer integration

The illustration of heterogeneous integration in figure 7 shows the transfer integrationapproach, which allows to transfer layers of one substrate to another substrate. Theselayers can be closed layers or devices which are fabricated in the same layer. The layerto be integrated is first fabricated on a temporary carrier substrate, which is the sourcesubstrate. The source substrate is then bonded upside down onto the target substrate,followed by removal of the source substrate. As a result, the source layer to beintegrated is bonded upside down onto the target substrate. If the element should notbe integrated upside down, it can be bonded to another intermediate substrate beforeit is transferred to the final target substrate. An example for transfer integrationis the integration of the thin device layer of a SOI wafer onto CMOS substrate formicromirror arrays.

In contrast to transfer integration, in direct integration the substrates are bondeddirectly on top of each other without intermediate carriers or removing substrates.The substrates can be bonded with the top or the bottom side facing each other,depending on the application. Direct integration allows for wafer-scale encapsula-tion/packaging of devices as illustrated in figure 10. The source substrate is bondedwith the top side onto the top side of the target substrate. The bonding layer isthick and patterned, forming a ring around the devices and encapsulating/packagingthe devices after the bonding. Encapsulating MEMS structures by bonding plainwafers with recesses on top of the target wafer is a common wafer-level packagingapproach [58]. However, in contrast to only one substrate containing devices, hetero-geneous integration allows for both of the substrates containing functional structures.

The integration approaches introduced above allow for integrating wafers to wafers,which is typically cost-efficient if the devices on both substrates feature similar foot-print areas or if the total cost per footprint of the substrate with the smaller devicesis much lower as compared to the final total cost of the final devices. Figure 11 il-lustrates this issue: if the device to be integrated features a much smaller footprintarea than the device on the target substrate, a lot of expensive substrate material isbe wasted.

The conventional alternative to integrate devices with largely different areas wouldbe the hybrid integration. However, this requires robotic pick-and-place of the com-ponents, which is a serial process and especially for high volume production may notbe cost-effective. Alternatives based on heterogeneous integration are chip to waferintegration methods, where the source substrate with the source devices is diced intosingle chips, which are which are bonded onto the target devices on the un-diced


encapsulated/packaged devices

bonding layer

top view

cross-section

rings

target devices

devices to be integrated

Figure 10. Schematic illustration of direct integration which results in wafer-level encapsu-lation/packaging of the integrated devices, based on a thick bonding layer which is patternedto form rings around the devices.

target wafer. These methods address the cost issue associated with largely differentfootprint areas and three examples of such techniques are described below.

One method is mixing pick-and-place and wafer-level transfer integration. Thesource substrate is diced into single chips, which are pick-and-place bonded onto thetarget devices on the target substrate. This approach has been developed by IMEC-Ghent University for the integration of optical chips onto IC substrates [59], using anintermediate substrate to bond the chips with the correct side to the target substrate.

Another approach utilizes self-alignment methods, which are part of self-assemblymethods [60, 61, 62, 63]. The smaller devices are densely fabricated on a source sub-strate which is diced into single chips. These single chips are placed on an assemblywafer, without any orientation or order and due to previous manipulations of the as-sembly wafer the chips orient themselves along defined patterns with defined pitches.Induced vibration or evaporating liquid helps to overcome the friction between thechips and the surface of the assembly wafer. After the chips have oriented and assem-bled themselves on the assembly wafer, they can be transferred onto the target wafer.Alternatively, the chips are self-assembled directly on the target wafer. Similar meth-ods have been used to transfer monocrystalline silicon onto surface micromachinedelectrostatic actuators to fabricate micromirrors [64]. Furthermore, there is a reportusing self-assembly for the integration of MEMS (actuators) with IC [65].

A third method is the ’selective transfer technology for microdevice distribution’presented by IBM [66]. Their concept involves the dense fabrication of the smallerdevices on a source wafer. This source wafer is aligned to the target wafer in away, that one small source device is aligned to the larger target device on the targetwafer. Then, the source device is released from the source wafer and transferredto the target device. By adapting the pitch of the smaller elements and the targetdevices, several smaller elements can be transferred in one transfer step. Using thistechnology, one source wafer can populate a number of target wafers and thereby the


source devices”wasted”

substrate area

target devices(a) (b)

Figure 11. Cost-effectiveness of wafer-to-wafer integration: (a) The integration is cost-effective because the footprint of the transferred device is similar to that of the target device.(b) The wafer-integration is no longer cost-effective due to the large footprint differences.

cost of the transferred wafer is distributed over the number of target wafers. IBMshowed this method for the distribution of their AFM-cantilevers and, together withFZK Karlsruhe, they showed this method for integrating bulk TiNi actuators ontopolymer microvalves [67].

3.2.2 Electrical interconnection

One of the advantages of heterogeneous integration is that the electrical intercon-nections between the integrated devices can be made very short and dense, reducingparasitic influences. There are applications, where electrical interconnection betweenthe integrated devices is not necessary (such as the integration of SMA, which isdescribed in later sections). Yet, if electrical interconnections are necessary (as forexample in the micromirror arrays), these interconnections are in most cases elec-trically conductive vertical vias. The methods for providing electrical vias can bedistinguished in the via first and the via last approach. In the via first approach,all vertical electrical interconnections are fabricated prior to integration and duringbonding the devices are electrically interconnected as illustrated in figure 12a. Oneof the advantages of this approach is that the two substrates can be fabricated com-pletely separately, are then electrically interconnected during the integration and nofurther post-integration processing is required for the electrical contacting. However,this method requires a careful alignment of the two substrates, which limits the sizereduction of the vias.

In the via last approach, only the electrical contact pads on the target substrate arefabricated prior to integration. After the bonding, vias are etched into the integratedsubstrate and filled with conductive material, as illustrated in figure 12b. In thisapproach, no precise wafer-to-wafer alignment is necessary and the vias can be madeconsiderably smaller than in the via first approach. However, the integrated substrate


electrical contacts

preparing electrical contacts prior to integration

the devices are electrically interconnected during the bonding

(a) via first

preparing electrical contacts only on target devices bonding of the substrates

etching of vias in the integrated devices and filling them with

conductive material(b) via last

substrate removed

substrate removedelectrical contactsvias

Figure 12. Illustration of the two approaches for electrical via fabrication: (a) the viafirst approach, where the connections are fabricated prior to integration and (b) the via lastapproach, where the connections are fabricated after integration.

must be processed after the bonding. As an example for the via last approach, thesilicon micromirrors in the mirror arrays (figure 9b) are electrically connected by viasetched through the silicon and the sacrificial layer to the electrical contact pads on theIC substrate. These vias are filled with metal to connect the mirrors to the drivingcircuitry and allow for electrostatic actuation of the mirrors.

3.2.3 Wafer-bonding techniques

A key technology of heterogeneous integration is the bonding of wafers to each other.The following brief descriptions are based on wafer bonding review papers [68,69]. Inprinciple, all the mentioned bonding methods are suitable for heterogeneous integra-tion.

In solder bonding [70, 71, 72], layers of metal or metal-alloy based solders areused to bond two wafers. The metal layers are usually deposited on both wafers,which are joined and heated to the melting temperature of the solder. The solderreflows and wets both wafer surfaces, causing intimate contact and bonding of thesurfaces. Example solder materials are lead-tin (Pb–Sn), gold-tin (Au–Sn) and tin-copper (Sn–Cu) solders. Oxides at the metal surfaces can result in poor bonding andtherefore most solder bonding processes use flux to remove the oxides. To some extent,solder bonding tolerates particles and structures at the wafer surfaces. The methodprovides hermetic bonding/packaging. Furthermore, it allows for combined bondingand vertical electrical interconnection, making it very interesting for heterogeneousintegration.

Eutectic bonding [73,74,75,76,77] is a variation of solder bonding, allowing to join


two wafers with dissimilar surface materials which form a eutectic mixture at tem-peratures much lower than their melting temperatures. The most common materialcombination is silicon (Si) and gold (Au) with a eutectic temperature of 363 °C. Eu-tectic bonding can result in strong and hermetic bonds at relatively low temperaturesand is therefore often used for the hermetic sealing of micromachined transducers.Furthermore, the method is interesting for heterogeneous integration because it al-lows for vertical electrical interconnection.

In adhesive bonding [69, 78, 79, 80], an intermediate adhesive layer creates a bondbetween two surfaces. Most commonly, a polymer adhesive is applied and the wafersare pressed together. Then, the polymer adhesive is hard-cured, typically by expos-ing to heat or ultraviolet (UV) light. The main advantages include the relatively lowbonding temperatures between room temperature and 450 °C (depending on the poly-mer material), the insensitivity (to some extent) to the topology or particles on thewafer surfaces, the compatibility with standard complementary metal-oxide (CMOS)semiconductor wafers and the ability to join practically any wafer materials. Whileadhesive wafer bonding is a comparably simple, robust and low-cost process, concernssuch as limited temperature stability and limited data about the longterm stabilityof many polymer adhesives in demanding environments need to be considered. Also,adhesive wafer bonding does not provide hermetically sealed bonds towards gassesand moisture. The method does not provide electrical interconnection.

In direct or fusion bonding [81,82,83], two wafers are contacted without significantpressure, electrical fields or intermediate layers. For reliable bonding, this methodrequires very flat and very clean wafer surfaces, room temperature contacting of thewafers and an annealing step (typically between 600 and 1200 °C) to increase thebond strength. This method results in strong and hermetic bonds and is thereforeof interest if the integration method should also provide hermetic packaging. Themethod does not provide electrical interconnection.

Anodic or field assisted bonding [84,85] is based on joining an electron conductingmaterial such as silicon and a material with ion conductivity such as alkali-containingglass. Heating to temperatures of 180–500 °C mobilizes the ions and an appliedvoltage of 200–1500 V creates a large electric field that pulls the wafer surfaces intointimate contact and fuses them together. Anodic bonding is more tolerant to surfaceroughness than direct bonding and usually leads to strong and hermetic bonds. Themethod is interesting if hermetic packaging is required, however, the large voltagesmight damage IC devices on the substrates. The method does not provide electricalinterconnection.

Thermocompression bonding, metal-to-metal direct bonding, and ultrasonic bond-ing [86, 87, 88, 89] are related bonding schemes in which two surfaces are pressedtogether and heated. Typically at least one of the surfaces consists of a metal. Thesurfaces plastically deform and fuse together. Instead of heating, the energy can alsobe supplied by ultrasonic energy (ultrasonic bonding), with the advantage of break-ing through native oxides, particles and surface nonuniformities at the bond interface.Common bonding surface materials are gold to gold, copper to copper, aluminum togold, and aluminum to glass. The disadvantage of thermocompression and ultrasonicbonding is that large net forces are required when bonding larger wafer areas. Thus,thermocompression bonding, metal-to-metal direct bonding and ultrasonic bonding


are mainly used in wire bonding schemes and in bump bonding schemes. However,these methods provide hermetic bonding/packaging and the possibility for verticalelectrical interconnection, making it very interesting for heterogeneous integration.

In low-temperature melting glass bonding [90] an inorganic low-temperature melt-ing glass or glass frit layer forms the intermediate bonding material and is depositedon one or both of the wafers. The wafers are joined and heated, causing the glassto deform or reflow and bonding the wafers. Two different types of glasses are avail-able; devitrifying glasses, of which the melting point is permanently increased afterthe curing and vitreous glasses, which always melt at the same temperature. Thismethod allows to hermetically bond various wafer materials at relatively low bond-ing temperatures and tolerates to some extent particles and structures at the wafersurfaces. The method does not provide electrical interconnection.

3.2.4 Releasing structures for actuation

When integrating actuators there are some more issues to consider beside wafer bond-ing and vertical electrical interconnection. Actuators imply moving structures whichmust be detached from their underlying bonding layer to allow their movement. Forstructures fabricated using wafer-to-wafer bonding, the techniques to detach themfrom their underlying substrate can be summarized in two approaches. The firstmethod is the localized bonding of areas to be affixed while avoiding the bonding ofthe structures to be detached. The second method is a bond-and-release approach,in which all structures are bonded to the substrate, followed by removing the bondinterface material underneath the structures to be detached.

Localized bonding (illustrated in figure 13a) between two substrates can be ob-tained using two different principles. The first principle is to modify the interfacematerial prior to bonding, defining bonding and non-bonding areas. Examples ofpatterned bond interface layers include adhesive layers applied only on areas wherebonding is desired [80] and bond blocking layers such as gold or platinum defining lo-cal non-bonding areas in anodic bonding [91]. The second localized bonding principleis to use heat triggered bonding methods and to localize the heat to the desired areasof the bond interface. Examples of this approach include integrated heaters for bothlocalized eutectic and silicon fusion bonding [92], localized soldering using inductiveheating [93] as well as local heating using lasers [94].

In localized bonding, the non-bonded parts are either fallout-structures or theymust remain mechanically connected to the bonded parts by mechanical supports toprevent them from falling out during the remaining process steps. The removal ofmechanical support structures through dicing or through controlled fracture has beenshown [95, 96]. However, such break-away structures limit the design freedom andpotentially increase the footprint area of the MEMS device. Furthermore, the movingstructures could be damaged while removing the support structures.

The most common technique for releasing bonded structures is the bond-and-release approach based on sacrificial underetching (illustrated in figure 13b). Thistechnique requires the fabrication of the structures on top of a ’sacrificial’ layer,which can be etched with a high selectivity. This approach is common in surface


bonded not bonded

to be detachedto remain bonded

(a) localized bonding of the structures to remain bonded

bonding layer

to remain bondedto be detached

(c) localized removal of the bonding layer

bonding layer is locally dissolved

bonding layer is not attacked

bonding layer

to remain bonded, without etch holes

to be detached, with etch holes underetch distance w

(b) sacrificial underetching with etch holes

w/2

Figure 13. Illustration of the different methods to detach structures from the substrate:(a) localized bonding of areas to be affixed while avoiding the bonding of the structures to bedetached; (b) sacrificial underetching with etch holes; (c) localized removal of the bondinglayer only underneath the structure to be released.

micromachining [97], where the layers are stacked upon each other. In the cross-bar switches presented later in this thesis, the actuators are fabricated using surfacemicromachining and released prior to integration with the target substrate. A widevariety of sacrificial materials have been demonstrated, such as silicon dioxide [98],polymers [99, 100] and metals [101]. In structures fabricated by wafer-bonding, sac-rificial underetching is based on wafer-bonding methods with intermediate bondinglayers that can be sacrificially etched with a high selectivity to release the attachedstructures. Examples of such intermediate bonding materials are silicon dioxide [102]and polymers [69, 103]. It should be noted that processes using the buried oxidein silicon-on-insulator (SOI) wafers as sacrificial layer can be considered as such abond-and-release technology.

A challenge in sacrificial underetching is to provide the selectivity between struc-tures to be detached and structures to remain bonded. If the whole sacrificial layeris removed, there is no selectivity and all structures are detached. One solution forthis issue is to affix the structures not to be detached via a second clamping material,which is not attacked when etching the sacrificial layer. An example for this strategyare again the silicon micromirror arrays: the silicon is transfer integrated using adhe-sive polymer bonding onto a polymer sacrificial layer. Then, metal vias which provideboth electrical and mechanical interconnection from the silicon to the IC substrateare fabricated (see figure 9b). Finally, all polymer is removed but the mirrors are stillclamped by the metal vias, which are not attacked during the polymer ashing.

Another approach which does not require a second clamping, is to provide the


selectivity by integrating vertical etch holes in the structures to be detached but notin the substrates to remain bonded. These etch holes drastically minimize the distanceto underetch as compared to structures without etch holes. Thus, the structures withetch holes are detached while the structures without etch holes are only partiallyunderetched at their edges (see figure 13b).

Etch holes are commonly included in sacrificial underetching to minimize the timethe whole substrate is exposed to the etchant or the etch process environment, whichmay result in device destruction if other materials than the sacrificial layer are at-tacked. Yet, beside the advantages there are also some challenges to consider. Etchholes potentially decrease the mechanical stability and the performance of the struc-tures to be released. Furthermore, the fabrication of such etch holes is feasible onlyfor structures consisting of thin layers, such as the thin silicon layers of micromirrorarrays. If the structures are hundreds of micrometers thick, etch holes are difficultto fabricate with the required aspect ratio. Even if such deep and thin vertical etchholes could be fabricated, the etching of the underlying layers would presumably bevery slow due to poor mass transport and depletion of the etchant in the etch holes.

An alternative, which allows for the area-independent release of bonded struc-tures without additional etch holes and harsh chemical environments, is the localizedremoval of the intermediate bonding layers (illustrated in figure 13c). An exampleof localized removal is the localized laser ablation of a polymer bonding layer [66].Another example is based on the electrochemical etching of metal sacrificial layers ina neutral salt solution. This principle has been shown for detaching surface microma-chined structures [104], which were deposited onto aluminum as sacrificial layer. Inthis thesis, this method was adapted to release eutectically bonded silicon microstruc-tures to allow for their manipulation by heterogeneously integrated SMA actuators.Further information about this release etch technique can be found in the attachedpaper 4.

4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS 27

4 Knife gate microvalves with bulk SMA microactu-ators

This chapter introduces methods for wafer-level integrating the shape memory alloyTitanium-Nickel (TiNi) and their combination with knife gate microvalves which area new tool for controlling large gas flows. First experiments proof the great promiseof the knife gate valve concept, however, the thermal bimorph actuator of the firsttest devices proved not robust enough and bulk TiNi material was identified as apromising replacement. There were no reported methods to wafer-level integratebulk TiNi actuators with silicon structures and, thus, no methods for combiningbulk TiNi with the knife gate microvalve concept. TiNi is difficult to integrate withconventional methods and therefore large parts of this thesis deal with developingmethods for heterogeneous integration of bulk TiNi material with silicon structuresand methods for cold-state reset and concepts for thermal energy supply.

The following sections will briefly introduce the challenges of integrating TiNimaterial with silicon structures and the concept of knife gate microvalves. Finally,first concepts to integrate bulk TiNi sheets and wires for actuating the valves will bepresented, together with initial measurements indicating an outstanding performanceof the TiNi actuated knife gate microvalves.

4.1 Integration of Titanium-Nickel shape memory alloy

Despite the advantages, the shape memory alloy Titanium-Nickel (TiNi) is not yet astandard MEMS actuator material, partially due to the difficulties to integrate TiNiwith silicon structures. So far, TiNi is integrated either monolithically or in a hybridfashion.

In monolithic integration, the TiNi is deposited directly onto the silicon structuresusing metal deposition techniques such as evaporation or sputtering. After deposition,the film is amorphous, i.e. it displays no shape memory effect (SME) and is com-pressively stressed. To functionalize the layer, the film is crystallized in an annealingstep, which involves heating the film to typically above 535 °C [105]. The annealingcan even be performed by heating the substrate during deposition [106, 13]. Duringcrystallization the stress in the film relaxes and after subsequent cooling the film isunder tensile stress, which can be utilized to form a bimorph actuator. In the coldstate, the silicon structure (typically a cantilever) is stiff enough to overcome the ten-sile stress and remains flat, thereby deforming the TiNi and providing the cold-state


(a) (b)

Figure 14. The thin SMA film actuated microgripper, consisting of two TiNi/siliconbimorphs which are eutectically bonded together: (a) a schematic illustration [107] and(b) a photograph of the device [21].

reset. Upon heating, the TiNi film contracts and forces the silicon cantilever to bend.Upon cooling the silicon cantilever flattens the bimorph again. A well known examplefor this technique is the microgripper [21, 107], which is shown in figure 14. In thebimorph approach, the cold-state reset is provided by the built-in stress between theTiNi and the silicon, thus, the cold-state reset is integrated monolithically with theTiNi.

Another approach is to remove the silicon underneath the TiNi, which results ina free standing TiNi film. This method requires additional elements to provide thecold-state reset. Commonly, such additional elements include a bias spring and amechanical spacer, with the spacer placed in between the bias spring and the TiNifilm. During the assembly of the parts, the bias spring deflects the TiNi via the spacer.When heated, the TiNi flattens and deflects the bias spring via the spacer. Theseelements are integrated either on wafer-level by bonding of other wafers containingthe cold-state reset elements or on device-level in a hybrid fashion. Figure 15 showsexamples for both cases [108,1].

Monolithic integration comes with several advantages, such as the wafer-level in-tegration of the material and the possibility of a built-in cold-state reset when usingthe bimorph approach. However, monolithic integration of TiNi also comes with somemajor challenges [21]. The transformation temperature of TiNi is very sensitive tovariations of Ti or Ni content. An increase in the atomic percentage of nickel by1% from an equi-atomic ratio of Ti:Ni lowers the martensitic start temperature from50 °C to -100 °C. Therefore the deposition technique must provide precise and reliablecontrol of the deposition rate of the two different metals, which is difficult to achieve.Even if a film has a proper global composition, local inhomogeneities result in the nu-cleation and growth of precipitates that inhibit shape memory behavior. The purityof the TiNi targets and high deposition vacuum are essential to limit impurities in thedeposited film, which also potentially inhibit the SME. The high temperature anneal-ing step causes compatibility problems with other materials and requires additionallayers to avoid unwanted interdiffusion between TiNi and silicon. Also, the large


(a) (b)

Figure 15. Schematic drawings of TiNi actuated microvalves based on the free-standingTiNi film approach with (a) silicon bias spring and spacer integrated via wafer-bonding [108]and (b) external, pick-and-place integrated metal bias spring and saphire ball spacer [1].

thermal stresses after deposition and during annealing could cause delamination orcracking of the films. Furthermore, the thicknesses of the films are limited to approx-imately 20 μm [5] and a recent report states that TiNi-based film sputtering is mostlyfeasible for thicknesses up to 10 μm only [109], potentially resulting in limited me-chanical robustness of structures actuated by TiNi films. For completeness it shouldbe mentioned that in a recent report [110] a 30 μm thick film of a ternary TiNi basedalloy was achieved by flash-evaporating TiNiCu onto copper substrates, however, theissues of complicated processing and high post-deposition annealing temperatures arestill not overcome.

One method to address these issues is the hybrid integration, where the TiNiis fabricated separately in dedicated factories and pick-and-place integrated as bulkmaterial on device level. Bulk TiNi is commercially available in different shapesand dimensions, for example wires with diameters starting from 18 μm and thinsheets starting from 20 μm (even thicknesses down to 5 μm have been reported [5]).Similar to the free-standing thin TiNi film approach, the cold-state reset is providedby additional elements, which are pick-and-place integrated on device level. As anexample, figure 16 shows a schematic drawing [111, 5] illustrating the principle ofhybrid TiNi integration for microvalves.

Hybrid integration of bulk TiNi comes with several advantages. The material isproduced with standard metalworking techniques in dedicated factories, resulting inspecified and reproducible material properties and thereby addressing the sensitivityto variations of Ti and Ni content in the alloy. However, the pick-and-place integra-tion of the TiNi in combination with the pick-and-place integration of the cold-statereset elements is difficult for microdevices and potentially results in large fabricationcosts for larger volume production.


Figure 16. Schematic illustration of hybrid integration of TiNi to fabricate a mi-crovalve [111, 5]. The bulk TiNi element is patterned and pick-and-place integrated withthe other parts, including the spacer for the cold-state reset. An applied pressure pushesthe membrane up, which via the spacer deforms the TiNi element. When heating, the TiNipushes the membrane down to the flow orifice, thereby closing the valve.

Heterogeneous integration methods allow for the combination of the two methodsby integrating bulk TiNi material and providing a cold-state reset on wafer-level.There is one report on heterogeneous TiNi integration [67], in which a SMA sheet waspatterned on wafer-scale and the elements were selectively transferred to single plasticmicrovalves using the selective transfer integration approach of IBM [66] mentionedabove. However, the cold-state reset was provided by a spacer between the microvalveand the SMA element, which again requires pick and place assembly. Large parts ofthis thesis deal with wafer-level integration of both TiNi and the cold-state reset.

4.2 Gas microvalves

4.2.1 Background

A valve is a device that regulates the flow of gas or liquids. In this thesis, the focuslies on millimeter sized valves for gas flow regulation. It would go beyond the scope ofthis thesis to review all the research efforts into microvalves, therefore the interestedreader is referred to review papers [112,113] for a more detailed overview.

Most of the active microvalves use mechanical moving parts which are coupled tointegrated actuators. Depending on the flow control principle, these microvalves canbe divided into diaphragm/seat type valves and gate type valves. Diaphragm/seattype valves comprise a boss (most common is a flexible diaphragm) which moves inparallel to the gas flow (see figure 17a) and which is coupled to an actuator. In theclosed state, the boss completely closes the orifice and the actuator must overcomethe pneumatic pressure, thus, . a strong actuator is necessary for controlling highpressure. To allow for a large flow in the open state, the distance and thereby theflow channel between the seat and the boss should be maximized, thus, the actuatormust provide a large stroke [114].


(b)

orifice

GateGuide

Guide

gas flow

(a)

orifice

gas flow

BossGui

de

Gui

deFigure 17. Illustration of: (a) seat valve with the flow regulating boss moving in parallelto the gas flow, requiring a strong and large-stroke actuator; (b) gate valve with the flowregulating gate moving perpendicular to the flow, reducing the requirements on the actuator.

In contrast to seat valves, the gas flow in gate valves is regulated by a gate whichmoves perpendicular to the gas flow (see figure 17b). In the ideal case, when fullyopening the valve the gate is completely retracted out of the flow channel and thereis no obstruction in the flow path, resulting in higher flow rates compared to a similarsized diaphragm/seat type valve. When moving the gate to close the flow path, thepneumatic pressure is counteracted by the mechanical guidance of the gate and in theideal case the actuator is only moving the gate and must not counteract the pneumaticpressure. However, the limited actuation energy available in microsystems does notallow friction between sliding structures and therefore requires spacing between floworifice and gate, which results in leak flow in the closed valve state. Fortunately, manyvalve applications tolerate leak flow.

Previously published work in the field of gate microvalves always featured gatesthat move in the wafer plane [115, 116, 117, 118, 119]. Figure 18 [1] shows schematicdrawings of a sliding plate gate valve [115], which is currently commercialized by thecompany Microstaq [120,121,122]. Yet, to allow the gate to move in the wafer-planerequires extra footprint area and increases the footprint-related costs.

The silicon footprint consumed by the valve is a good cost indicator and whencomparing microvalves to each other not only the performance in terms of controlledflow and pressure should be considered, but also the performance obtained per con-sumed footprint area. Therefore, in this thesis the comparison figure ’controlledflow/pressure drop/footprint’ is suggested with the unit [sccm/kPa/mm2]. Table 4shows the performance per footprint of different microvalves and the numbers indicate,that gate microvalves typically have a much larger performance than diaphragm/seatmicrovalves.

Microvalves should be fabricated cost-efficiently to allow them to be successfuloutside the specific niche markets they currently focus on. This thesis presents theknife gate microvalve concept, in which the gate moves out of the wafer plane tominimize the chip footprint area and the footprint-related cost.


Figure 18. Schematic drawings of the sliding gate microvalve introduced by Williams etal [115], showing (a) a cross-section and (b) a three-dimensional drawing of the three siliconwafers, which are assembled to form a pressure-balanced microvalve. The figure is copiedfrom [1].


Table 4. Comparison of different valves in terms pneumatic performance relative to theirconsumed footprint area. It should be noted, that not all of the listed valves are siliconvalves, however, they are all microvalves. The performance comparison is related to theknife gate valve presented in this work.

Reference and actuation method Q P w l Q/P/A

related tovalve in the

attachedpaper 1

this work, knife gate microvalvesthermal bimorph, attached paper 1,externally actuated

3400 95 2.3 2.3 4.21 100%

SMA (TiNi sheet) actuation 1000 8 6.5 6.5 2.95 70%SMA (TiNi wire) actuation 3800 50 2.0 4.5 8.44 200%

other gate microvalves[115], thermal 6700 1000 2.5 5.0 0.54 13%[116], thermal 5000 100 4.0 4.0 3.13 74%[119], externally magnetic 500 40 1.4 1.4.0 6.17 147%

seat microvalvesZdeblick et al [123], phase change,(’FLUISTOR’ valve)

5000 140 6.0 6.0 0.99 23.59%

TiNi Alloy company, SMA (TiNi),figure 15, [1]

1000 690 8.0 5.0 0.04 0.86%

Messner et al [124], thermal bimorph 1150 800 4.0 4.9 0.09 2.14%Jerman [125], thermal bimorph 100 170 2.5 2.5 0.09 2.24%Fu et al [126], magnetic 2000 200 10.0 10.0 0.10 2.38%Yang et al [127], electrostatic 45 900 1.64 1.64 0.02 0.44%Li et al [128], PZT 39 210 20.0 20.0 0.00 0.01%Yang et al [129], PZT 52 2070 8.4 5.0 0.00 0.01%Shao et al [130], PZT 70 50 13.0 13.0 0.01 0.20%Huff et al [131], electrostatic 800 550 3.6 3.6 0.11 2.67%Barth et al [132], thermal bimorph 1000 344 8.8 8.8 0.04 0.89%Yanget al [133], thermopneumatic 1000 228 8.0 8.0 0.07 1.63%Bosch et al [134], magnetic +electrostatic

3 16 3.0 8.0 0.01 0.19%

Kohl et al [135, 5], SMA (TiNi) 1600 120 20.0 11.0 0.06 1.44%Kohl et al [136, 5], SMA (TiNi) 470 500 6.0 6.0 0.03 0.74%Kohl et al [137, 5], SMA (TiNi) 180 210 6.0 11.0 0.01 0.31%Kohl et al [137, 5], SMA (TiNiPd) 360 250 6.0 11.0 0.02 0.52%Kohl et al [138, 5], SMA (TiNiCu) 360 300 6.0 11.0 0.02 0.43%Kohl [5], SMA (TiNi) 280 200 6.0 6.0 0.04 0.92%Kohl et al [139, 5], SMA (TiNi) 460 800 6.0 6.0 0.02 0.38%

Q = flow in sccm; P = pressure in kPa; w,l = width, length in mm; A = w×l = footprint in mm2

Q/P/A = performance per footprint in sccm/kPa/mm2


gas gas

(a) front gate valve

gas (b) side gate valve (c) back gate valve

actuator

gate

gate

gate

actuator actuatorgas

channelgas

channelgas

channel

Figure 19. Illustration of the three different knife gate microvalve designs introduced indetail in the attached paper 1.

4.2.2 Knife gate microvalves

This thesis presents the knife gate microvalve concept, in which the gate moves outof the wafer plane to minimize the chip footprint area and the footprint-related costs.Figure 19 illustrates the three different designs of microvalves with the gate mov-ing out-of-plane. The gate is coupled to a monolithically integrated thermal sili-con/aluminum bimorph actuator and moves out-of-plane to regulate an in-plane flow.The valve concept is explained in detail in the attached paper 1.

Figure 20 confirms the valve’s potential for controlling large flows: an active chiparea of only 2.3×3.7 mm2 allows for a flow change of ΔQ = 3400 standard cubiccentimeters per minute (sccm) at a pressure difference of 95 kPa.

Table 4 lists the performance per footprint of different microvalves. For simplercomparison the presented knife gate microvalve is chosen as reference and all theother microvalves are compared to it. The comparison shows that the knife gatemicrovalve provides the highest performance per footprint, only outperformed by arecently published rotary gate valve and the TiNi wire actuated knife gate microvalve.The rotary valve is electromagnetically actuated using an external magnetic field andbelongs therefore to another class of microvalves with external actuation mechanisms.

leak flow from im-perfect assembly

leak flow from gatespacing

pressure [kPa]20 40 60 80

flow

[scc

m]

1000

2000

3000

4000

5000

0

100% open

50% open

closed

Figure 20. Measured pressure-flow characteristics of a back-gate microvalve. The dashedline indicates the expected leak caused by imperfect assembly.


(a) (b) (c)

TiNi sheet actuator

gate

carrier substrate

eutectic bonding layer

polymer bonding layer

locally remove bonding layer

basegas flow

Figure 21. Cross-sectional drawing of the first design of a TiNi sheet actuated knife gatevalve showing (a) a back gate valve bonded to a carrier substrate and the TiNi sheet priorto integration; (b) the adhesive bonding of the TiNi sheet onto the valve and (c) releasingthe flow regulating gate to allow for its manipulation by the actuator.

The TiNi wire actuated knife gate microvalve is a successor of the thermal bimorphknife gate microvalve and will be discussed below.

However, it should be noted that while the experiments indicate the excellentcapabilities of the knife gate microvalve concept, the thermal bimorph actuator failedfor several reasons. First, the bimorph beam was not strong enough to withstandthe vibration and torsional forces caused by the flow. Second, improper thermaldesign caused improper heating and actuation of the beam. Bulk TiNi actuators wereidentified as a promising replacement, however, concepts for integrating bulk TiNimaterials had to be developed first.

4.3 TiNi sheet actuated knife gate valves

4.3.1 Design

In a first, yet unpublished design to integrate TiNi sheets onto silicon knife gatemicrovalves, the TiNi is integrated with a back gate valve (back gate valve see fig-ure 19c). Figure 21 shows the design, in which the back-gate valve is bonded ontoa carrier substrate with a vertically through-etched hole, providing an enclosed flowchannel which is expected to simplify the pneumatic connection. The TiNi is bondedonto the valve: one end of the TiNi is bonded to the flow channel forming the basefor the cantilever and the other end is bonded onto the flow-regulating gate.

4.3.2 Fabrication - Integration of TiNi sheets

A method was investigated to adhesively bond TiNi sheets onto silicon structures, to-gether with methods for the cold-state reset and thermal energy supply. The attachedpaper 2 contains a detailed description of these methods.

The concept to integrate TiNi sheets onto silicon structures is illustrated in fig-ure 22a. The TiNi sheet is patterned to form the actuator structures, using flexureinterconnections to hold them together. The patterning is performed prior to integra-tion to avoid the exposure of the silicon target structures to the very aggressive TiNietchant. The bonding adhesive is stamped onto the surface of the patterned target


flexures holding cantilevers together

bonding to target waferaligning to target waferpatterning of TiNi sheet

wafer- sized, bulkTiNi sheet

target wafer coated with adhesive layer

etched holes to allow for cantilever movement

(a) TiNi sheet integration concept

hot statecold state

resistive heater

stress layer

bulk TiNi

(b) trimorph actuator concept

Figure 22. Illustration of (a) final bulk TiNi sheet integration concept and (b) trimorphactuator concept, as developed in paper Nr. 2.

wafer, which was a ’dummy’ wafer with etched holes to allow the TiNi actuators todeflect freely. The stamping method is specially developed in this thesis and basedon standard dicing blue tape. Then, the TiNi sheet is aligned and bonded onto thetarget wafer. For developing this integration method, the target wafer was a ’dummy’wafer with etched holes, which allow the TiNi actuators to deflect freely.

After the TiNi material is integrated, the cold-state reset and thermal energysupply must be provided for actuation. In this work, these elements are integratedmonolithically with the TiNi sheet in form of three functional layers. Figure 22billustrates the actuation concept. The first functional layer is the core actuation layerformed by bulk TiNi material. The second functional layer is a stressed film whichdeforms the actuator in the cold-state. This film is deposited onto the TiNi andeliminates the need for pick-and-place integration of an additional bias spring. Thethird functional layer is a heating layer to supply thermal energy to the actuator viaan indirect heating scheme and to avoid the complicated electrical contacting of theTiNi alloy.

In the valve design shown in figure 21, the TiNi sheet is bonded onto the gate,which itself is bonded to another silicon wafer. To allow the gate to be manipulated forcontrolling the flow, its underlying bonding layer must be removed without removingthe bonding layers underneath the other silicon structures. Section 3.2.4 presents somemethods for the localized release of structures while keeping other structures bonded,


0 5 10 15

200

600

1000

1400

1800

pressure [kPa]

flow

[scc

m]

open state

closed state

Figure 23. First measurements of the pressure-flow characteristics of a yet unpublishedTiNi-sheet actuated knife gate microvalve.

however, most of them require underetching of the structures to be released, includingexposure of the whole device to harsh environments and etching of all bonding layers.To avoid these issues, a new bond-and-release method was developed: The valveand the carrier substrate are bonded together using eutectic Au-Si bonding, followedby localized removal of the eutectic bonding layer. For a detailed description andinvestigation of this method, the reader is referred to the attached paper 4.

4.3.3 Results

Figure 23 shows first results of the flow performance of a gate microvalve with TiNisheet actuation. An active chip area of only 6.5×6.5 mm2 allows for a flow change ofΔQ ≈ 1000 sccm at a pressure difference of 8 kPa. These numbers result in a flowperformance per footprint (see Table 4) of about 70% of the first knife gate valve withthe non-functional thermal bimorph actuators. However, in contrast to the first knifegate valve, the TiNi sheet actuated valve was actuated with the integrated actuatorand did not require any external actuation. Furthermore, the tested TiNi sheet knifegate valve is an early prototype with room for optimization.

4.3.4 Discussion

The TiNi sheets are integrated directly onto the silicon structures, without any inter-mediate carrier or transfer substrate. Since the sheet is patterned prior to integration,it has to be aligned to the target wafer. Both alignment and placing onto the adhesivelayer were performed manually. However, the patterned TiNi sheet was fragile andvery easily deformable, which complicated the handling. The alignment of the sheetwas enabled only by a high ambient temperature caused by a hotplate underneaththe target wafer, which triggered the stiff hot state of the sheet. To allow for analignment and bonding using a standard wafer alignment and bonding tool, futuredesigns should aim for a transfer integration method, in which the SMA sheet is pre-processed on an intermediate carrier substrate and then transfer bonded from thecarrier substrate onto the target wafer.


TiNi cantilever

eutectic Au-Sibonding area

silicon

probe needles

TiNi cantilever with stress layer

heater

(a) (b)

Figure 24. (a) Photographs of a eutectically bonded TiNi cantilever which was severelyplastically deformed by manual pulling with a tweezer without breaking the bond, therebyillustrating the extreme bond strength. (b) Picture of a eutectically bonded TiNi trimorphactuator with a resistive heater.

The sheets are affixed onto the silicon structures using adhesive polymer bonding.The adhesion strength looks promising, however, more tests are necessary to verify thebond stability, especially for long-term cycling of the actuator. An alternative methodcould be the one used for the micromirror arrays presented in section 3: the adhesivebonding could serve as temporary mechanical fixation and electroplated metal viasthrough the TiNi sheets could provide both electrical interconnection and permanentmechanical fixation. Yet another alternative is to use a different bonding methodwhich allows for direct mechanical fixation. Figure 24a shows non-published picturesof first experiments testing the eutectic bonding of TiNi to silicon and illustrating theextreme bond strength: a eutectically bonded TiNi cantilever was severely plasticallydeformed by manual pulling with a tweezer without breaking the bond.

At this stage, an electrical interconnection between the TiNi and the silicon struc-tures is not necessary, since any scheme for electrical contacting of the actuator wouldmost likely involve wire bonding directly to the metal heater on top of the TiNi tri-morph actuators. Since the work in the attached paper 2 was focused on integrationand cold-state reset, the test structures did not contain a metal heater. Figure 24bshows a photograph of a yet unpublished complete trimorph actuator containing aresistive metal heater, which is eutectically bonded to silicon. The heater was con-tacted using probe needles and consumed approximately 100 mW to trigger the TiNihot state.

4.4 TiNi wire actuated knife gate valves

4.4.1 Concept

Figure 25 shows a first, yet unpublished design to integrate TiNi wires with the knifegate valve concept. The wires are integrated with a front gate valve (front gate valvesee figure 19c).


(a) (b) (c)

TiNi wire

gas flow

gate

polymer bonding

Figure 25. Cross-sectional drawing of the first design of a TiNi wire actuated knife gatevalve showing (a) a front gate valve and the TiNi wires prior to integration; (b) the adhesivebonding of the TiNi wires onto the valve and (c) the actuation.

4.4.2 Fabrication - Integration of TiNi wires

For the SMA wire actuators, the integration concept is derived from the actuationconcept, which is illustrated in figure 26a. TiNi wires are strained by pulling themprior to integration. Then, the strained wires are integrated with silicon cantileversby anchoring one end of the wires to the base of the silicon cantilever and the otherend of the wire to the moveable end of the cantilever. Upon heating, the TiNi wiresrecover the strain, i.e. they contract and thereby force the silicon cantilever to bendout of plane. In the cold state, the silicon cantilevers act as cold-state reset, recoveringtheir flat shape by stretching the TiNi wires.

Figure 26b illustrates the resulting wafer-level integration concept. The wires areoriented and placed on a dedicated tensioning frame, which is used to strain thewires. Then, the wafer and the wires are aligned to each other and put in contact.Finally, the wires are anchored for each actuator and the tensioning frame is removed.After integration, the wires are diced together with the silicon chips. For a detailedinvestigation including the proof of concept by fabricating first test structures thereader is referred to the attached paper 3. The fabricated test devices showed strokesof up to 396 μm for a cantilever length of 3 mm, which is among the highest values incomparison to different microactuators reported in a recent review paper [6]. Sincethe work in this paper was focused on integration and cold-state reset, no heatingconcept is presented.

4.4.3 Results

Figure 27 shows first results of the flow performance of a knife gate microvalve withTiNi wire actuation. An active chip area of only 2×4.5 mm2 allows for a flow changeof ΔQ ≈ 3800 sccm at a pressure change of 50 kPa. These numbers result in aflow performance per footprint (see Table 4) of about 200% of the first knife gatevalve with non-functional thermal bimorph actuators. In contrast to the first knifegate valve, the TiNi wire valve is actuated with the integrated actuator. The flowperformance per consumed footprint area is a world record compared to the othervalves and clearly demonstrates the promise of TiNi actuated knife gate microvalves.

4.4.4 Discussion

The TiNi wires are transfer-integrated, using a dedicated tensioning frame as tempo-rary carrier. For the first test devices, the alignment was performed manually and thewires were placed onto the adhesive anchors using a dedicated lift stage. This process


straining wire cold state hot state

silicon

TiNi wirestrained TiNi wireanchor

anchor

operational states of the actuator

anchoring

polymer anchors

straining wiresclamping wires

TiNi wires

target wafer with silicon cantilevers

fixed clamp moveable clamp

aligning and contacting

(a) actuator concept

(b) wafer-level integration concept

Figure 26. Illustration of (a) the actuation concept using TiNi wires on silicon cantileversand (b) concept for wafer-level integration of the wires, as developed in the attached paper 3.

0 10 20 30 40 500

2000

4000

6000

pressure [kPa]

flow

[scc

m]

leakage due to imperfect assembly

theoretical flow lines

open state

closed state

Figure 27. First measurements of the pressure-flow characteristics of a yet unpublishedTiNi-wire actuated knife gate microvalve.


valvewirebondingelectrical connector pins

pneumatic connector tube

TO 8 housing

Figure 28. Illustration of the packaging concept for the knife gate microvalves. The valve ispackaged in a standard TO8-housing, like it is used for pressure sensors. The Picture showsa pressure sensor of the Robert Bosch GmbH with the lid removed.

could be adapted to implement a high precision (x,y,z) stage, thus, the process lookspromising for the wafer-level integration of wires. However, prior to integration thewires need to be placed onto the tensioning frame with the correct orientation andpitch, which was performed manually in this work. Future work should verify thatthe wires can be placed onto the tensioning frame using automated methods.

The wires are attached to the silicon cantilevers using anchors made of SU-8, whichis locally cured with UV light. However, the area underneath the wires is shadowedby the wires and therefore not fully cured. Future work should implement reflectivepatterns in the silicon underneath the wires to reflect the UV light into the SU-8volume underneath the wires. Alternatively, the wires could be anchored using metalclamping via electroplating metallic connections between the silicon and the wires.

4.5 OutlookThe work presented above is a big step towards a microvalve with potential for com-mercial application. However, a fully integrated TiNi actuated valve with integratedheaters has not been shown so far. All necessary elements have been discussed andfuture work will focus on combining all elements to show a fully functional TiNiactuated microvalve.

For commercialization the valve must be packaged. Figure 28 shows the suggestedpackaging concept, which is inspired by the packaging of several commercial pressuresensors. The valve is glued onto the pneumatic connector tube of a TO8-housing,thereby pneumatically interconnecting the valve to an inlet pressure. The outlet ofthe flow can be provided by a hole in the lid of the TO8-housing. The electrical inter-connection to the actuator on the valve is provided by wirebonding to the electricalinterconnection pins of the TO8-housing.

5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPEDACTUATOR SWITCHES 43

5 Automated main distributing frames with S-shapedactuator switches

This chapter presents a MEMS crossbar switch as the core unit to automate maindistributing frames, in which the parties in copper-wire telecommunication networksare cross-connected. The crossbar switch consists of an array of MEMS switcheswhich are individually addressable and allow to interconnect a number of input linesto a number of output lines in any combination desired. In this work, the switchesare based on the electrostatic S-shaped actuator described earlier in section 2.3.2.Each of the switches in the array is individually addressable using a row/columnaddressing method, avoiding a complicated CMOS circuitry as utilized in micromirrorarrays. The fabrication concept of the crossbar switch is based on the basic conceptof heterogeneous integration, which involves the separate fabrication of parts followedby their assembly to a device.

The following subsections briefly introduce automated main distributing frames,followed by a description of the S-shaped actuator switch. Finally, the MEMS crossbarswitch and the concept of individual switch addressing are presented.

5.1 Switch units in automated main distributing frames

The current telecommunication networks evolved from the original telephone net-works and they are for the most part still based on copper-wire networks. Withinthese networks, the main distributing frame (MDF) is the center of cross-connectingequipment of service providers to their customers and it is located both in large centraloffices and remote street cabinets. The MDF still relies upon the same routines andequipment it did 50 years ago. Hence, whenever a cross-connection must be changed,this re-configuration must be carried out manually by technicians physically chang-ing the interconnection wires (jumpering) in the MDF. Before the deregulation of thetelecommunication markets the customers only rarely changed their services and therewere no or little economical incentives to automate the jumpering. Nowadays, how-ever, the customers are highly flexible and there is a growing need to adapt quickly tonew telecommunication services. Due to the increased labor-intensive manual jumper-ing the network maintaining costs are drastically rising and thereby creating a clearincentive to automate the jumpering using an automated MDF (AMDF).

The core element of the AMDF is a crossbar switch. Originally, the term ”crossbar


+- +- +-

+-

+-

+-

crossbar switch unit

inpu

t cha

nnel

s

output channels

double switch

inpu

t cha

nnel

s output channels (b) multicast switching networks (a) electrical diagram of

a switch unit

Figure 29. Electrical and hierarchical schematic diagrams: (a) switch unit, consisting ofan array of 3 × 3 double switches; (b) arrangement of switch units on higher hierarchicallayers using the Clos method to minimize the necessary amount of switches.

switch” relates to a grid of metal bars crossing each other without touching each otherand which are electrically interconnected or isolated by switches in their crosspoints.In a AMDF, the metal bars are formed by signal channels, with a set of input channelsand a set of output channels. A channel consists of a signal line pair and switchingthe signal line pair simultaneously requires a double-pole single-throw (DPST) switchin each cross-point. Typically, the cross-bar switches are symmetrical, with N inputsand N outputs resulting in N2 crosspoints. Figure 29a shows an electrical schematicof the signal circuit for a simplified crossbar switch with N=3. A typical AMDFconnects tens of thousands of channels and it is not feasible to provide this amount ofcrosspoints in a single crossbar switch since it is difficult to fabricate such large arrayswith an acceptable yield. Furthermore, each bar or channel can only be connected inone crosspoint while the other crosspoints remain unused and therefore large parts ofthe crossbar switch are not utilized. A solution is to fabricate small crossbar switchunits with a low N and interconnect several of them to a bigger crossbar switchunit, which itself again forms a switch unit in a bigger crossbar switch. In 1953,Charles Clos at Bell labs developed the ’minimal spanning switch’ method [140] tointerconnect the small crossbar switch units in a way to reduce the total number ofswitches and interconnections, but still allowing for a non-blocking network whichalways provides one route to interconnect an input to an output. In this method, theswitch units are arranged in three stages with each the same number of switch unitsand each switch unit sharing only one connection with a switch unit in another stage,as illustrated in figure 29b. If a path is blocked, the connections can be reroutedvia other switch units to open a new path. When scaling up the resulting crossbarswitch, Clos’ interconnection method reduces the amount of crosspoints from thequadratically scaling N2 to the more favorable 3×√

N×√N×√

N = 3×N1.5, therebydrastically minimizing the number of switches and the related costs. As an example,a conventional crossbar switch with N = 100 would require N2= 10000 crosspointswitches, whereas a Clos crossbar switch requires 10 switch units à 10×10 crosspoints


Figure 30. Illustration of the switch unit in a robot based AMDF: The signal channelsare interconnected in their crosspoints by inserting/extracting metal pins using a robot ona xy-stage. The figure is copied from [145].

in three stages, which results in 10×10×10×3 = 3000 crosspoint switches, only.It would go beyond the scope of this thesis to extend this introduction, therefore

the interested reader is referred to literature such as [140,141,142,143,144].

A crossbar switch unit comprises three basic elements. The first element is therouting network with input and output signal channels crossing each other withouttouching each other. The second element are switches needed for interconnecting thesignal channels in their crosspoints. The third element is controlling of the switches:each switch must be individually addressable to allow for the interconnection of anyof the N input lines to any of the N output lines in any configuration desired.

To ensure the highest flexibility when connecting new services and new technolog-ical equipment to the customers, the signal path should ideally act like a copper-wirecable which requires ohmic mechanical switching with metal contacts and excludessolid-state relay based crossbar switching. Previous attempts to provide a switchunit for a Clos network in an automated MDF consist of robots [145, 146, 147] orelectromagnetic relays [148,149].

Figure 30 [145] exemplarily illustrates the robot approach: the routing network isformed by a matrix board and the switches are metal pins, which are inserted/extractedin the crosspoints of the signal channels using a robot. The addressing is provided bymoving the robot to the addressed crosspoint using a xy-stage with a laser for finealignment. In the electromagnetic relay approach, the crossbar switching is performedby an array of electromagnetic relays on printed circuit boards. These arrays includecontrol circuits with diodes and transistors to allow for a row/column addressing.The attached paper 5 contains a short comparison of the technologies, which did notprovide solutions with the desired performance at acceptable costs

Nowadays, MEMS technology offers the possibility to integrate a large number ofmicro-switches on a single chip and utilizes high-volume semiconductor manufactur-


(a) OFF

switchingcontact

stiff cantilevertop electrode

bottomelectrode large gap between

electrodes

(b) ON

Figure 31. Illustration of a conventional cantilever-type electrostatic MEMS switch. Thefigure is modified from [37].

ing methods and logistics which potentially results in low production costs. MEMSswitches have been investigated extensively during the last decade for many applica-tions requiring high quality signal switching [150,151,152]. They fill the gap betweensolid-state relays (SSR) and electromagnetic relays (EMR) by offering true ohmicswitching, high miniaturization and integration and very good signal properties overa large bandwidth. To the authors’ knowledge, there are no scientific reports onMEMS crossbar switches or switch arrays for the described application. However,there are a few reports on small-scaled and unpackaged metal-contact MEMS switcharrays for other applications [153,154,155,156,157,158].

5.2 MEMS switches

5.2.1 Background

MEMS switches are devices that mechanically open or short-circuit a transmissionline. They are of submillimeter size and they outperform semiconductor switches interms of power consumption, on-state resistance, cut-off frequency, isolation and signalinsertion loss [150,151,152]. There are two basic types of MEMS switches: capacitiveswitches and metal-contact switches. Capacitive switches are typically used as shuntswitches between the signal line and the ground line while metal-contact switches areused as series switches between two transmission lines, opening or closing the signalpath. In metal-contact switches, the signal is conducted through metal contacts andin capacitive switches, the contact is made between the metal and dielectric material.The application in automated MDF’s described above requires ohmic switching ofsignals from DC to max. 50 MHz, therefore only metal contact series switches are ofinterest in this work.

For metal contact switches there are several requirements to consider when de-signing the switch. In the off-state, the switch should provide a high isolation whenapplying signals with high voltages, hence, a large distance between the switch con-tacts is necessary. When the switch is closed, the contact resistance should be aslow as possible and the contact should remain stable during the whole ON time. Theswitch contact resistance and reliability is directly influenced by the contact force andthe contact material. The influence of the contact materials is a research field on itsown [159,160,161,162] and not touched upon in this work. The contact force shouldbe maximized, however, a strong contact force increases the risk of stiction between


Figure 32. Illustration of the connection between the design parameters and the activeforces in the switch model and how they affect its performance in terms of contact resistanceand reliability. The figure is copied from [37].

the contacts. Therefore, the forces opening the switch should be strong enough toovercome the adhesion forces between the switch contacts.

Most of the MEMS switches are electrostatically actuated and the conventionaland so far most common electrostatically actuated MEMS switch is based on a simplecantilever approach, as illustrated in figure 31. The switch contact bar at the move-able end of the cantilever interconnects two signal lines when the cantilever is pulleddown, as illustrated in figure 31b. This configuration is basically a parallel-plateactuator and comes with all the problems mentioned in section 2.3.1, especially incombination with the switch requirements described above: a large distance betweenthe switch contacts for a high OFF state signal isolation results in a large initial elec-trode gap and thereby in large actuation voltages. To overcome stiction forces whenopening the switch, a stiff cantilever is required to provide strong opening forces.As a consequence, increased actuation voltages are necessary to overcome the strongspring force. Figure 32 [37] illustrates all the parameters involved for electrostaticallyactuated metal contact switches and how they are interconnected.

5.2.2 The S-shaped actuator switch

To overcome the limitations of the conventional cantilever type switch, an alterna-tive approach was developed based on the S-shaped film actuator described in sec-tion 2.3.2 [35, 36, 37]. The assembly concept of the S-shaped actuator highlights thisactuator for the heterogeneous integration with other structures. One substrate con-tains all the moveable parts of the actuator, while the target substrate provides anactuation electrode and a spacer to provide the gap in which the membrane movesup and down. Figure 33 illustrates the basic concept of the S-shaped film actuator


switch substrate

target substrateV

switch contact

signal lines to connect

(a) before assembly (b) after assembly: OFF (c) ON

d

Figure 33. The S-shaped actuator switch: (a) before assembly, (b) after assembly in theOFF-state and (c) in the ON-state.

switch. The switch substrate contains the flexible membrane with a switch contactand the target substrate contains the signal lines which are to be interconnected bythe switch (figure 33a). The distance d between the switch contacts in the OFF state(figure 33b) can be tuned and a large distance allows to obtain a high off-state isola-tion. In the ON state (figure 33c), the signal lines are interconnected by the switchcontact on the membrane. The switch contact area is surrounded by the actuationelectrode, resulting in a more homogeneous contact force.

If the spring force of the thin rolling membrane is not strong enough to open theswitch, the upper zipper-actuator provides an active opening capability.

5.3 The MEMS crossbar switch

The following subsections introduce the MEMS crossbar switches developed in thisthesis. These crossbar switches are fully fabricated using standard MEMS technolo-gies and materials. Thus, strictly speaking no heterogeneous integration is appliedsince heterogeneous integration refers to the integration of different technologies ormaterials. However, the development of the MEMS crossbar switch is based on thebasic concept of heterogeneous integration; the device is divided into substructures,which are fabricated separately and finally integrated with each other.

For the MEMS crossbar switch, the elements to integrate are a signal channelrouting network, switches to interconnect the signal channels in their crosspoints anda control line network for the individual addressing of single switches. The followingsections briefly introduce the routing network, the switches and their integration toa crossbar switch. Finally the addressing concept will be presented. A detaileddescription of the concept and the fabrication can be found in the attached paper 5and 6.

5.3.1 The routing network

Figure 34 (bottom part) illustrates the signal line routing network. The signal linesare fabricated in two different metal layers, which are isolated from each other by anintermediate polymer layer. The input lines are in the lower metal layer and the out-put lines are in the upper metal layer, crossing the input lines. In each crosspoint theinput lines are connected to the upper metal layer by vertical metal vias through the


spacer ring and adhesive layer

intermediate isolation layer

crosspoint of input and output

double switch in the crosspoints

vertical via through intermediate isolation

layer

signal -signal +electrode

output + bottom electrode on upper metal layer

signal +

signal -input on lower

metal layer

membrane electrodetop

electrodeassembly

assembly

bottom part with signal line routing network

top part with switches to interconnect signal lines

in their crosspoints

Figure 34. Three-dimensional drawing showing the concept of the KTH MEMS crossbarswitch, simplified to a 2 × 2 array. Some components of the bottom part have been removedto reveal details. The bottom part contains the signal-routing network on two metal layersand the bottom electrodes for the switch actuation. Furthermore, the picture shows the twoBCB layers for isolating the metal layers of the routing network from each other and forcreating the encapsulating spacer ring between the top and the bottom part. The top partcontains the moving part of the S-shaped film actuator switches with the switch contacts tointerconnect the signal lines in their crosspoints.


intermediate polymer layer. The upper metal layer also contains actuation electrodes(bottom electrodes) to actuate the switches later on.

5.3.2 The crosspoint switches

Figure 34 (top part) illustrates the switching part of the crossbar switch, which allowsto interconnect the signal lines in each crosspoint. The switches are S-shaped actuatorswitches as introduced above. As illustrated in the electrical schematic of the signalcircuit in figure 29a, the signal lines are paired to channels, consisting of a signal+and a signal− line. To switch both lines simultaneously, each membrane contains twoseparate switch contacts to allow for the parallel switching of two line pairs in thetarget substrate.

5.3.3 The integration

Finally, the top part with the switches is placed onto the bottom part with the signalchannels. The spacing between the two parts is provided by a polymer ring (figure 34,spacer ring) around the crosspoints. Besides spacing, the polymer ring is also designedto provide the mechanical fixation of the two parts. The polymer ring is patternedprior to the assembly, but not fully cured. After the assembly, the polymer is curedand thereby the two parts are adhesively bonded. Furthermore, the switches areencapsulated/packaged by the polymer ring, based on the encapsulation/packagingconcept illustrated in figure 10.

5.3.4 Individual switch addressing

There are several possibilities for the individual addressing of the switches. Theconventional approach would be to connect each switch with an individual controlwire. However, the number of switches and thereby the number of control wires scalequadratically with the size of the matrix, making this conventional approach feasibleonly for very small arrays with a small number of elements. Another approach isto integrate a CMOS addressing circuitry as shown earlier for micromirror arrays,which allows for the analog control of the actuators in the array as well as for highaddressing frequencies [50]. The presented fabrication concept of the MEMS crossbarswitch would allow for the heterogeneous integration of the switches with a CMOSwafer, however, the AMDF application requires only digital mode addressing withlow addressing frequencies and does not justify the integration of high-cost CMOScircuits.

The addressing approach developed for the MEMS crossbar switch utilizes thehysteresis behavior of the actuation voltages of electrostatic actuators, which wasexplained in section 2.3.1 and illustrated in figure 3c. Applying a set voltage Vset

which is larger than the pull-in voltage always pulls in the moveable plate and theswitch is ON: Vset ≥ Vpull−in. Reducing the applied voltage to a hold voltage Vhold

within the hysteresis range between the pull-in and the pull-out voltage holds themoveable plate in the current position, i.e. either in the ON or in the OFF state:


Vhold Vhold Vhold Vhold Vhold Vhold

-Vset-

Vset++

Vhold Vhold Vhold

Vhold + Vset+ - (-Vset-) > Vpull-in

Vhold - (-Vset-) < Vpull-in

Vhold + Vset+ < Vpull-in

Vpull-out < Vhold < Vpull-in

Vpull-out < Vhold < Vpull-inVpull-out < Vhold < Vpull-in

(a) (b) (c)

Figure 35. Schematic drawings illustrating the individual setting of cell (2,1) in a 3×3array and the voltages involved. (a) Hold state. Each actuator maintains its current state.(b) Pull-in of the actuator by applying the addressing voltages at the respective rows andcolumns. (c) Return to hold state. Maintaining the programmed states of all cells.

Vpull−out < Vhold < Vpull−in. To open the switch, the applied voltage is reduced tothe reset voltage Vreset, which is smaller than the pull-out voltage: Vreset ≤ Vpull−out.

For the addressing scheme to work, the electrodes of the fixed plates are inter-connected in each respective column, whereas the electrodes of the moveable platesare interconnected in each respective row. This arrangement allows to provide therequired potential differences in the crosspoints of the electrodes and the numberof necessary control lines is reduced from the quadratically scaling N2 to the linearscaling complexity of 2N. As an example, with the conventional addressing approacha crossbar switch with N = 20 would require 400 control lines, while the presentedrow/column approach requires only 2N = 40 control wires.

The individual pull-in is illustrated in figure 35. If all columns are fed with thehold voltage Vhold and all rows are on ground potential, all switches maintain theircurrent states (figure 35a). To set the actuator at the address (2,1), the potentialdifference between the electrodes must exceed the pull-in voltage Vpull−in . Therefore,a voltage level of Vhold + Vset+ is applied at the 2nd row, and a negative voltage levelof Vset− is applied at the 3rd column (figure 35b). After the individual pull-in thevoltages ΔVset+ and ΔVset− are removed and the columns are connected to the holdvoltage Vhold, maintaining the current state of all switches as illustrated in figure 35c.

The individual pull-out is illustrated in figure 36. All columns are fed with the holdvoltage Vhold and all rows are on ground potential, all switches maintain their currentstates (figure 36a). To reset the actuator at the address (2,1), the potential differencebetween the electrodes must be below the pull-out voltage Vpull−out. Therefore, apositive voltage level of Vreset is applied at the 3rd column and thereby subtracted


Vhold Vhold Vhold Vhold Vhold Vhold

Vreset

Vhold Vhold Vhold

Vhold - Vreset < Vpull-out

Vhold + Vreset - Vreset = Vhold

(a) (b) (c)


Vreset+

Vreset+

Vhold +Vreset < Vpull-in



Figure 36. Schematic drawings illustrating the individual resetting of cell (2,1) in a 3×3array and the voltages involved. (a) Hold state. Each actuator maintains its current state.(b) Pull-out of the actuator by applying the addressing voltages at the respective rows andcolumns. (c) Return to hold state. Maintaining the programmed states of all cells.

from Vhold. To avoid resetting all the switches in the 3rd column, Vreset must beadded to Vhold at the other rows (figure 36b). Finally, all columns are connected tothe hold voltage Vhold, maintaining the current state of all switches as illustrated infigure 36c. Another simpler method for the resetting is to connect all electrodes toground and set a new pattern as described above.

This addressing scheme has been reported before for micromirror arrays [163,164,165,166,167] and programmable capacitor banks [168]. Neither of these reports, how-ever, discusses any design requirements or considerations on the operational robust-ness of the addressing scheme. This thesis contains a full investigation for determiningthe operational reliability of the addressing scheme and to optimize the addressingvoltage levels, taking into account the statistical deviations of the actuation voltagesboth for a single actuator as well as across the array. The attached journal paper 6extensively investigates this issue and presents methods to choose actuation voltageswhich result in a statistical certainty of better than 99.9% for the addressing to work.

5.4 Discussion

The routing network and the switches are integrated directly, without any interme-diate carrier or transfer substrate. The mechanical fixation of the two substrates isdesigned as adhesive bonding via the patterned adhesive spacer ring. In this design,the polymer spacer ring not only provides spacing and mechanical fixation, but alsoan encapsulation and thereby packaging of the crosspoints. However, it should benoted here, that even though the mechanical fixation of the two substrates is de-signed as a wafer-level process, the first test structures were assembled manually on


Figure 37. Picture of the electrical contacting to evaluate the crossbar switch. The arrowindicates the probe needle from the bottom side to contact the electrodes on the top substrateof the crossbar switch.

chip-level using epoxy droplets in the corners for mechanical fixation. This issue willbe addressed in more detail below.

Usually, a controlled atmosphere inside the packaged volume is desired and there-fore hermetic sealing of the encapsulation would be the optimum. The polymer basedsealing of the presented crossbar switches does not provide hermetic sealing even ifthe chosen polymer, Bencocyclobutene (BCB) from the Dow Chemical Company, isoften referred to as ”near-hermetic” packaging [169]. A full hermetic packaging isobtained using a metallic bonding scheme, however, at the cost of the relatively non-complicated method to provide the spacing between the substrates using a polymerring.

The two substrates interact with each other both electrically and mechanically.Yet, there are no vertical electrical interconnection vias which allow to access theswitch control electrodes on the switch substrate via the network substrate. Thelack of this vias is the main reason, why the first test structures were assembledmanually on chip-level: both the switch substrate (top part) and the network substrate(bottom part) contain switch electrodes with contact pads. For the assembly theswitch substrate is flipped over and placed onto the spacer on the network substrate.Now the electrode contact pads of the switch substrate cannot be accessed from thetop anymore and so far there are no vertical electrical interconnection vias allowing tolink the electrodes on the switch substrate to contact pads on the network substrate.The chosen contacting method is illustrated in figure 37: the device is assembled onchip-level with an overlap between the rims of the two substrates and the contactpads are accessed from the top and from the bottom using probe needles. For theearly prototypes, this method was sufficient to evaluate the feasibility of the crossbarswitches, but future devices should contain vertical electrical interconnections.


Figure 38. Picture showing a MEMS crossbar switch after the assembly. A weight wasnecessary to force down the top part, counteracting the spring forces of the membranes, asillustrated in the left-hand inset. Then, the top part with the switches was aligned to thebottom part and finally the two substrates were fixated with epoxy droplets in the corners.

The moveable parts of the switches are released prior to integration: the mem-branes are fabricated by plasma enhanced chemical vapor deposition (PECVD) ofsilicon nitride on a polymer sacrificial layer. Assisted by etch holes in the membrane,the polymer was removed in a plasma ashing step. As a consequence, the thin mem-branes are not mechanically supported during the integration of the switch substrate,potentially damaging the membranes. However, the manual assembly has shown thatthe membranes of the switches are surprisingly mechanically robust. As an example:during the manual assembly, the top part with 400 silicon nitride membranes with athickness of 1 μm was placed on the bottom part. Because of large stress gradients inthe membranes the deflection of the free ends were much larger than the thickness ofthe spacer ring and the top part had to be forced down onto the spacer ring using aweight as shown in figure 38. In this state, the switches were manually aligned (align-ment accuracy ±5 μm), with considerable friction between the tips of the strainedmembranes and the surface of the bottom part. Despite the friction and the resultingstrains and stresses, in most cases all the membranes survived the procedure.

The stress gradients in the membranes are high, resulting in high actuation volt-ages especially for the 20×20 array (∼ 95 V) with shorter membranes and smallerelectrode area. Therefore the PECVD process should be optimized to allow for morecontrol over the stress gradients in the deposited films. Furthermore, when depositingthe metal film for the membrane electrode additional stresses are induced between themetal and the membrane. Also, the devices got very hot during the removal of thesacrificial layer underneath the membranes in an oxygen plasma . The combination ofheat and plasma influences the stresses in the silicon nitride membranes [170,171,172],which issues should be further investigated.

Most of the switches failed to provide a good ohmic contact. Assumed reasons are


that the contact force was not high enough and the surfaces of the switch contactswere contaminated since the switch was not assembled in a cleanroom environment.In future devices, the contact force can be increased by lowering the stress-gradientin the membrane and by local stiffening of the membrane to improve the transfer ofthe electrostatic force between the surrounding electrodes to the switch contacts. Toavoid contamination, future devices should be assembled in the cleanroom, ideally onwafer-level and after a short plasma cleaning of the surfaces.

The actuators were very susceptible to stiction, which was most probably causedby electrostatic stiction between the electrodes. In general, electrostatic stiction isan issue for electrostatic actuators where dielectric layers are involved. Charges areinjected and trapped in the dielectric layers, thereby causing an electrostatic attrac-tion. In the MEMS crossbar switch, charges can be trapped in the silicon nitride ofthe membranes and the electrical isolation layers as well as in the large amount ofBCB underneath the bottom electrodes. Consequently, in future devices the amountof dielectric materials should be kept at a minimum.

5.5 OutlookCurrently a second generation of MEMS crossbar switches is under development,which features arrays of 20×20 double-switches on a silicon footprint of only 12×12 mm2.In the revised fabrication no complicated electroplating of thick metal layers is nec-essary and the overall stress gradients in the membranes can be controlled using astress-compensation design. The chip is fully hermetically packaged by metallic wafer-bonding and contains vertical electrical interconnection between the two substratesas well as through-silicon vias through the network substrate to allow for hybrid inte-gration as introduced in section 3. To obtain a more reliable operation of the switchesall polymers are removed to reduce the risk of electrostatic stiction caused by trappedcharges.

6 SUMMARIES OF THE APPENDED PAPERS 57

6 Summaries of the Appended Papers

Paper 1: Out of plane knife gate microvalves for controlling large gas flowThis paper considers design issues for microvalves for large gas flow control. Itintroduces out-of-plane knife-gate microvalves as a novel design concept and aproportional microvalve concept for pressure control applications. The design ofthree different actuator-gate configurations and first prototypes are presented.The first valve prototypes feature thermal silicon–aluminum bimorph actua-tors and the pressure-flow performance per chip area of the demonstrator valvepresented is greatly increased using out-of-plane actuation and an out-of-planeorifice. The characterization of the actuators and of the pressure-flow perfor-mance is presented.

Paper 2: Wafer-scale manufacturing of bulk shape memory alloy microactuators basedon adhesive bonding of Titanium-Nickel sheets to structured silicon wafersThis paper presents a concept for the wafer-scale manufacturing of microactu-ators based on the adhesive bonding of bulk shape memory alloy (SMA) sheetsto silicon microstructures. Wafer-scale integration of a cold-state deformationmechanism is provided by the deposition of stressed films onto the SMA sheet. Aconcept for heating of the SMA by Joule heating through a resistive heater layeris presented. Critical fabrication issues were investigated, including the cold-state deformation, the bonding scheme and related stresses and the Titanium-Nickel (TiNi) sheet patterning. Novel methods for the transfer stamping of ad-hesive and for the handling of the thin TiNi sheets were developed, based on theuse of standard dicing blue tape. First demonstrator TiNi cantilevers, wafer-level adhesively bonded on a microstructured silicon substrate, were successfullyfabricated and evaluated. Intrinsically stressed silicon dioxide and silicon nitridewere deposited to deform the cantilevers in the cold state and the resulting tipdeflections were evaluated.

Paper 3: Design and wafer-level fabrication of SMA wire microactuators on siliconThis paper reports on the fabrication of micro actuators through wafer-levelintegration of pre-strained SMA wires to silicon structures. In contrast toprevious work, the wires are strained under pure tension, and the cold statereset is provided by single crystalline silicon cantilevers. The fabrication isbased on standard MEMS manufacturing technologies and enables an actuationscheme featuring high work densities. A mathematical model is discussed, which


provides a useful approximation for practical designs and allows analyzing theactuators performance. Prototypes have been tested and the influence of con-structive variations on the actuator behavior is theoretically and experimentallyevaluated. The test results are in close agreement with the calculated values,and show that the actuators feature displacements among the highest reported.

Paper 4: Localized removal of the Au-Si eutectic bonding layer for the selective re-lease of microstructuresThis paper presents and investigates a novel technique for the footprint andthickness-independent selective release of Au–Si eutectically bonded microstruc-tures through the localized removal of their eutectic bond interface. The tech-nique is based on the electrochemical removal of the gold in the eutectic layer andthe selectivity is provided by patterning the eutectic layer and by proper elec-trical connection or isolation of the areas to be etched or removed, respectively.The gold removal results in a porous silicon layer, acting similar to standardetch holes in a subsequent sacrificial release etching. The paper presents theprinciple and the design requirements of the technique. First test devices werefabricated and the method successfully demonstrated. Furthermore, the paperinvestigates the release mechanism and the effects of different gold layouts onboth the eutectic bonding and the release procedure.

Paper 5: Single-chip MEMS 5×5 and 20×20 double-pole single-throw switch arraysfor automating telecommunication networksThis paper reports on microelectromechanical (MEMS) switch arrays with 5×5and 20×20 double-pole single-throw (DPST) switches embedded and packagedon a single chip, which are intended for automating main distribution framesin copper-wire telecommunication networks. Whenever a customer requests achange in his telecommunication services, the copper-wire network has to bereconfigured which is currently done manually by a costly physical re-routingof the connections in the main distribution frames. To reduce the costs, newmethods for automating the network reconfiguration are sought after by the net-work providers. The presented devices comprise 5×5 or 20×20 double switches,which allow us to interconnect any of the 5 or 20 input lines to any of the 5 or 20output lines. The switches are based on an electrostatic S-shaped film actuatorwith the switch contact on a flexible membrane, moving between a top and abottom electrode. The devices are fabricated in two parts which are designed tobe assembled using selective adhesive wafer bonding, resulting in a wafer-scalepackage of the switch array. The on-chip routing network consists of thick metallines for low resistance and is embedded in bencocyclobutene (BCB) polymerlayers.

Paper 6: Row/Column addressing scheme for large electrostatic actuator MEMSswitch arrays and optimization of the operational reliability by statistical analy-sisThis paper investigates the design and optimization of a row/column address-ing scheme to individually pull in or pull out single electrostatic actuators inan N2 array, utilizing the electromechanical hysteresis behavior of electrostatic

6 SUMMARIES OF THE APPENDED PAPERS 59

actuators and efficiently reducing the number of necessary control lines from N2

complexity to 2N. This paper illustrates the principle of the row/column ad-dressing scheme. Furthermore, it investigates the optimal addressing voltagesto individually pull in or pull out single actuators with maximum operationalreliability, determined by the statistical parameters of the pull-in and pull-outcharacteristics of the actuators. The investigated addressing scheme is imple-mented for the individual addressing of cross-connect switches in a microelec-tromechanical systems 20×20 switch array, which is utilized for the automatedany-to-any interconnection of 20 input signal line pairs to 20 output signal linepairs. The investigated addressing scheme and the presented calculations weresuccessfully tested on electrostatic actuators in a fabricated 20×20 array. Theactuation voltages and their statistical variations were characterized for differ-ent subarray cluster sizes. Finally, the addressing voltages were calculated andverified by tests.

7 CONCLUSIONS 61

7 Conclusions

This thesis presents methods for the wafer-level integration of TiNi and S-shaped elec-trostatic actuator switches to functionalize MEMS devices. The integration methodsare based on concepts of heterogeneous integration. Using the developed integrationmethods, TiNi actuators are integrated with a new type of microvalves, drasticallyimproving their performance. The microvalves are knife gate microvalves and in-troduced in this thesis. The S-shaped actuator switches are arranged in arrays andintegrated with a signal line routing network, forming a MEMS crossbar switch.

In more detail, the developed TiNi integration methods:

– allow to wafer-level integrate bulk TiNi sheets and wires, together with mecha-nisms for the cold-state reset, avoiding the complicated monolithic integrationand the pic-and-place hybrid integration. First actuator test structures werefabricated and evaluated, showing promising results. Heating with integratedheaters is shown for the sheet based actuators;

– provide techniques for handling the TiNi prior to integration. The TiNi sheetsare patterned prior to integration to avoid exposure of the target structures tothe very aggressive TiNi etchant. The thin TiNi wires are oriented and strainedbefore they are transferred to the target substrate.

– are discussed, including initial tests of alternative approaches.

The developed knife gate microvalves:

– feature a gate moving perpendicular to the flow, reducing the demands on theactuator. The gate moves out-of-plane to control an in-plane gas flow, minimiz-ing the required silicon footprint area and the related costs;

– display excellent pneumatic performance in relation to the consumed siliconfootprint area, however, the first test structures required external actuationsince the integrated actuators failed;


For combining knife gate microvalves and TiNi actuators, the thesis shows:

– first designs for integrating TiNi sheets and wires with silicon knife gate mi-crovalves, including a method for the localized release of eutectically bonded,thick silicon microstructures to allow their manipulation by the integrated ac-tuators;

– first experiments displaying outstanding pneumatic performance of the valve inrelation to the consumed silicon footprint area.

– a discussion and outlook on future work, including a scheme for packaging ofthe valve.

The MEMS crossbar switch:

– is intended as core switch unit for automating parts of copper-wire telecommu-nication networks;

– consists of two parts, with one part containing signal input and output linescrossing each other and the other part containing an array of S-shaped actuatorswitches. After integrating the switch array on the signal lines using adhesivebonding, the switches allow to interconnect the input to the output lines in anycombination desired;

– features integrated encapsulation of the switch arrays;

– contains up to 20×20 = 400 crosspoint switches on an area of 12×16 mm2.First prototypes were fabricated and evaluated, proofing the feasibility of theconcept;

– requires a method for the individual addressing of selected switches. The thesispresents the concept of row/column addressing, drastically reducing the neces-sary amount of control wires. The reliability of the addressing is maximizedby determining the addressing voltages based on the statistical deviation of theactuation voltages across the array.

– is discussed and an outlook is given, based on current developments of a secondgeneration of MEMS crossbar switches.

7 CONCLUSIONS 63

(a) (b) (c)

Figure 39. Figure illustrating my gratitude to family, friends and colleagues in threedifferent languages: (a) Swedish, (b) English and (c) German.

Acknowledgments

This work was financially supported by Pondus Instruments AB, by Network Automa-tion mxc AB, by Vinnova (the Swedish Agency for Innovation Systems) through theSummit and the Forska&Väx framework, and by the European Commission throughthe sixth framework programme funded project Q2M.

As all big projects, the project ’PhD’ requires the support of so many other people.Figure 39 expresses what I want to say to you:

First of all, my principal supervisor Professor Göran Stemme for allowing me tostart the PhD-adventure in his group, yet always being there to limit the adventurepart and giving advice when needed. I truly appreciate that your door was alwaysopen to discuss major and minor issues, not only during my ups but mainly duringmy downs.

My supervisors Wouter van der Wijngaart and Joachim Oberhammer for showingme how the academic world and research works. Special thanks to Wouter for notonly being my supervisor, but for the feeling of being welcome in the early days andfor being a great friend who helped during those downs.

Frank Niklaus for sorting out the worst in the integration section of this thesisand for always having an open-minded approach to things.

Björn Samel for being a true friend, the great atmosphere in our ’German office’and for giving me the assurance that I’m not the only German fighting against Swedishwindmills. Thank you for our friendship !

Another good friend, Niklas Sandström, for showing me the Swedish way of lifeand of course for all the wonderful company when sharing hotel rooms (not like itsounds) and when enjoying our ’matlåda’ together.

Henrik Gradin for being a great room-mate and for the very fruitful work we didtogether. You’re a smart one and I hope some of the smartness diffused to my sideof the office.

Kjell Norén for his shared interest in everything with motors and wheels andfor demonstrating what a real engineer is. Thanks for all your brilliant ideas andprototypes which helped so much.

Petra Palmquist for proofreading this thesis without despairing over it.Special thanks also to my former professors, Peter Pokrowsky and Patrick Klär

for the support in the early days of my PhD adventure and for the remaining goodcontact.

Of course a big ’thank you’ also to all the other colleagues I had the pleasure


to work with in my projects. Sjoerd Haasl for the guidance and great spirit in myearly PhD days. Prof. Manfred Kohl, Dr. Thomas Grund and Johannes Barth fromFZK Karlsruhe for our fruitful work in the Q2M project and for the honor of invitingme as speaker to your seminar. Special thanks to Thomas and the ’Deutsche Eiche’Johannes for the nice time beside those meetings and conferences. Dr. Jan Peirs andDonato Clausi from KUL Leuven for the fruitful work in the Q2M project. Donato- I am glad I met you and I hope we will remain friends. Thorbjörn Ebefors, GöranEdström, Göran Lindvall and Jochen Walter from Silex Microsystems for the closecollaboration when developing the MEMS crossbar switch 2.0. Cecilia Aronsson forintroducing me to the cleanroom routines and to the Swedish language. SveinbjörgIngvarsdottír, Jutta Müntjes and Gaspard Pardon for enduring my guidance duringyour master thesis and placements.

Thanks to Erika Appel and Hanne Eklund for paving the way through KTH-bureaucracy and enabling things.

All the work was performed in a great atmosphere, for which I want to thankmy colleagues from KTH-MST: Göran, Frank, Joachim, Wouter, Hans, Erika, Kjell,Niclas, Mikael, Mikael, Mikael (yes, it’s a common Swedish name), Niklas, Fredrikand Carl Fredrik, Adit, Farizah, Andreas, Martin, Kristinn, Gaspard, Umer, Zargham,Hitthesh, Henrik, Nutapong, Staffan.

Besides work, there is also a personal life. I would like to thank all my friendswho helped me during this time by either distracting me from work or cheering meup in hard times. You know who you are - Thank you !

My family - always there for me. Thank you to my parents Regina and Kurtfor demonstrating the values of honesty and reliability and how to be almost perfectparents. I have not forgotten that you should be among the research funders in thefirst paragraph of the acknowledgments. Thank you to my brother Andreas for beingmy big brother. Thank you to my niece Charlotte for the smile when I think of you.Last, but not least - the biggest thank you to my Love Juliane for loving me the wayI am.

Das war meine Thesis - das Spiel geht weiter ...

REFERENCES 65

References[1] N. Maluf and K. Williams, An introduction to microelectromechanical systems

engineering, 2nd ed. Boston : Artech House, 2004.

[2] G. T. Kovacs, Micromachined Transducers Sourcebook. McGraw-Hill, 1998.

[3] E. Thielicke and E. Obermeier, “Microactuators and their technologies,” Mecha-tronics, vol. 10, no. 4-5, pp. 431 – 455, 2000.

[4] M. J. Madou, Fundamentals of Microfabrication, 2nd ed. CRC Press LLC,2002.

[5] M. Kohl, Shape Memory Microactuators. Springer, 2004.

[6] D. Bell, T. Lu, N. Fleck, and S. Spearing, “MEMS actuators and sensors: Ob-servations on their performance and selection for purpose,” Journal of Microme-chanics and Microengineering, vol. 15, no. 7, pp. 153 – 164, July 2005.

[7] S. A. Wilson, R. P.-J. Jourdain, Q. Zhang, R. A. Dorey, C. R. Bowen, M. Wil-lander, Q. U. Wahab, M. Willander, S. M. Al-hilli, O. Nur, E. Quandt, C. Jo-hansson, E. Pagounis, M. Kohl, J. Matovic, B. Samel, W. van der Wijngaart,E. W. Jager, D. Carlsson, Z. Djinovic, M. Wegener, C. Moldovan, E. Abad,M. Wendlandt, C. Rusu, and K. Persson, “New materials for micro-scale sen-sors and actuators: An engineering review,” Materials Science and EngineeringR: Reports, vol. 56, no. 1-6, pp. 1 – 129, 2007.

[8] D. P. Arnold and N. Wang, “Permanent Magnets for MEMS,” Journal of Mi-croelectromechanical Systems, vol. 18, no. 6, pp. 1255–1266, December 2009.

[9] J. Comtois, V. Bright, and M. Phipps, “Thermal microactuators for surface-micromachining processes,” Proceedings of the SPIE - The International Societyfor Optical Engineering, vol. 2642, pp. 10 – 21, 1995.

[10] S. Prasanna and S. Spearing, “Materials selection and design of microelec-trothermal bimaterial actuators,” Journal of Microelectromechanical Systems,vol. 16, no. 2, pp. 248–259, April 2007.

[11] D. Hodgson and J. Brown, “Using Nitinol alloys,” Shape Memory Applications,Inc., White paper, 2000. [Online]. Available: www.sma-inc.com

[12] F. Wang and W. Buehler, “Additional unique property changes observed duringTiNi transition,” Applied Physics Letters, vol. 21, no. 3, pp. 105 – 6, August1972.

[13] E. Wibowo and C. Kwok, “Fabrication and characterization of sputtered NiTishape memory thin films,” Journal of Micromechanics and Microengineering,vol. 16, no. 1, pp. 101 – 8, January 2006.

[14] W. Buehler, J. Gilfrich, and R. Wiley, “Effect of low-temperature phase changeson the mechanical properties of alloys near composition TiNi,” Journal of Ap-plied Physics, vol. 34, no. 5, pp. 1475 – 1477, May 1963.

www.sma-inc.com


[15] K. Otsuka and C. Wayman, Shape Memory Materials. Cambridge UniversityPress, 1998.

[16] Z. Wei, R. Sandstrom, and S. Miyazaki, “Shape-memory materials and hybridcomposites for smart systems: Part I shape-memory materials,” Journal of Ma-terials Science, vol. 33, no. 15, pp. 3743 – 3762, 1998.

[17] D. Stöckel, “Status and trends in shape memory technology,” in ProceedingsACTUATOR, 1992, pp. 79–84.

[18] Y. Bellouard, “Shape memory alloys for microsystems: a review from a materialresearch perspective,” Materials Science & Engineering: A (Structural Materi-als: Properties, Microstructure and Processing), vol. 481-482, pp. 582 – 9, May2008.

[19] H. Scherngell and A. Kneissl, “Training and stability of the intrinsic two-wayshape memory effect in Ni-Ti alloys,” Scripta Materialia, vol. 39, no. 2, pp. 205– 212, 1998.

[20] ——, “Generation, development and degradation of the intrinsic two-way shapememory effect in different alloy systems,” Acta Materialia, vol. 50, no. 2, pp.327 – 341, 2002.

[21] P. Krulevitch, A. Lee, P. Ramsey, J. Trevino, J. Hamilton, and M. Northrup,“Thin film shape memory alloy microactuators,” Journal of Microelectrome-chanical Systems, vol. 5, no. 4, pp. 270 – 82, December 1996.

[22] J. Van Humbeeck, “Shape memory alloys: a material and a technology,” Ad-vanced Engineering Materials, vol. 3, no. 11, pp. 837 – 50, November 2001.

[23] G. Eggeler, E. Hornbogen, A. Yawny, A. Heckmann, and M. Wagner, “Structuraland functional fatigue of NiTi shape memory alloys,” Materials Science andEngineering A, vol. 378, pp. 24 – 33, 2004.

[24] G. Song, “Robust position regulation of a shape memory alloy wire actuator,”Proceedings of the Institution of Mechanical Engineers. Part I: Journal of Sys-tems and Control Engineering, vol. 216, no. 3, pp. 301 – 308, 2002.

[25] N. Troisfontaine, P. Bidaud, and P. Dario, “Control experiments on two SMAbased microactuators,” Experimental robotics V (Lecture Notes in Control andInformation Sciences), vol. 232, pp. 490–499, 1998.

[26] N. Ma, G. Song, and H.-J. Lee, “Position control of shape memory alloy actu-ators with internal electrical resistance feedback using neural networks,” Smartmaterials and structures, vol. 13, pp. 777–783, 2004.

[27] B. Selden, K. Cho, and H. Asada, “Segmented shape memory alloy actuatorsusing hysteresis loop control,” Smart Materials and Structures, vol. 15, pp. 642–652, 2006.

REFERENCES 67

[28] M. Sreekumar, M. Singaperumal, T. Nagarajan, M. Zoppi, and R. Molfino,“Recent advanves in nonlinear control technologies for shape memory alloy ac-tuators,” Journal of Zheijang University SCEINCE A, vol. 8, no. 5, pp. 818–829,2007.

[29] J. Yao, “RF MEMS from a device perspective,” Journal of Micromechanics andMicroengineering, vol. 10, no. 4, pp. 9 – 38, December 2000.

[30] E. Hung and S. Senturia, “Extending the travel range of analog-tuned electro-static actuators,” Journal of Microelectromechanical Systems, vol. 8, no. 4, pp.497–505, 1999.

[31] L. Rocha, E. Cretu, and R. Wolffenbuttel, “Using dynamic voltage drive in aparallel-plate electrostatic actuator for full-gap travel range and positioning,”Journal of Microelectromechanical Systems, vol. 15, no. 1, pp. 69–83, 2006.

[32] C. G. Agudelo, M. Packirisamy, G. Zhu, and L. Saydy, “Nonlinear control of anelectrostatic micromirror beyond pull-in with experimental validation,” Journalof Microelectromechanical Systems, vol. 18, no. 4, pp. 914 – 923, 2009.

[33] J. Oberhammer, A.-Q. Liu, and G. Stemme, “Time-efficient quasi-static algo-rithm for simulation of complex single-sided clamped electrostatic actuators,”Journal of Microelectromechanical Systems, vol. 16, no. 2, pp. 373 – 382, 2007.

[34] M. Shikida, K. Sato, and T. Harada, “Fabrication of an S-shaped microactua-tor,” Journal of Microelectromechanical Systems, vol. 6, no. 1, pp. 18–24, March1997.

[35] J. Oberhammer and G. Stemme, “Low voltage high-isolation DC-to-RF MEMSswitch based on an S-shaped film actuator,” IEEE Transactions on ElectronDevices, vol. 51, no. 1, pp. 149–155, January 2004.

[36] ——, “Design and fabrication aspects of an s-shaped film actuator based dc torf mems switch,” Journal of Microelectromechanical Systems, vol. 13, no. 3, pp.421 – 428, 2004.

[37] J. Oberhammer, “Novel RF MEMS switch and packaging concepts,” Ph.D. dis-sertation, KTH-Royal Institute of Technology, 2004.

[38] O. Brand, “Microsensor integration into systems-on-chip,” Proceedings of theIEEE, vol. 94, no. 6, pp. 1160 – 76, June 2006.

[39] G. Fedder, R. Howe, T.-J. K. Liu, and E. Quevy, “Technologies for cofabricatingMEMS and electronics,” Proceedings of the IEEE, vol. 96, no. 2, pp. 306 – 22,February 2008.

[40] J. Smith, S. Montague, J. Sniegowski, J. Murray, and P. McWhorter, “Em-bedded micromechanical devices for the monolithic integration of MEMSwith CMOS,” International Electron Devices Meeting. Technical Digest (Cat.No.95CH35810), pp. 609 – 12, 1995.


[41] S. Lewis, S. Alie, T. Brosnihan, C. Core, T. Core, R. Howe, J. Geen, D. Hol-locher, M. Judy, J. Memishian, K. Nunan, R. Paine, S. Sherman, B. Tsang,and B. Wachtmann, “Integrated sensor and electronics processing for > 108

"iMEMS" inertial measurement unit components,” in Electron Devices Meeting,2003. IEDM ’03 Technical Digest. IEEE International, 2003, pp. 39.1.1–39.1.4.

[42] J. R. Hui, W. Xingtao, and W. C., “Adressing large arrays of electrostaticactuators for adaptive optics applications,” in Proceedings of the SPIE - TheInternational Society for Optical Engineering, vol. 5553, no. 1, October 2004,pp. 17–27.

[43] R. Tummala, “SOP: what is it and why? A new microsystem-integration tech-nology paradigm-Moore’s law for system integration of miniaturized convergentsystems of the next decade,” IEEE Transactions on Advanced Packaging, vol. 27,no. 2, pp. 241 – 9, May 2004.

[44] S. Al-Sarawi, D. Abbott, and P. Franzon, “A review of 3-D packaging technol-ogy,” IEEE Transactions on Components, Packaging and Manufacturing Tech-nology, Part B: Advanced Packaging, vol. 21, no. 1, pp. 2 – 14, February 1998.

[45] D. M. Jang, C. Ryu, K. Y. Lee, B. H. Cho, J. Kim, T. S. Oh, W. J. Lee, andJ. Yu, “Development and evaluation of 3-D SIP with vertically interconnectedthrough silicon vias (TSV),” Electronic Components and Technology Conference,pp. 847 – 52, 2007.

[46] R. Beica, C. Sharbono, and T. Ritzdorf, “Through silicon via copper electrode-position for 3D integration,” 58th Electronic Components and Technology Con-ference, pp. 577 – 83, 2008.

[47] J. Gu, W. Pike, and W. Karl, “A novel capillary-effect-based solder pump struc-ture and its potential application for through-wafer interconnection,” Journalof Micromechanics and Microengineering, vol. 19, no. 7, pp. 074 005 (7 pp.) –,July 2009.

[48] T. Bauer, “First high volume via process for packaging and integration ofMEMS / CMOS,” Whitepaper. [Online]. Available: www.silexmicrosystems.com

[49] S. Braun, “Praxissemesterbericht/Practical term report,” FachhochschuleKaiserslautern/Zweibrücken, Robert Bosch GmbH, Tech. Rep., 2001.

[50] H. Lakner, P. Durr, U. Dauderstaedt, W. Doleschal, and J. Amelung, “Designand fabrication of micromirror arrays for UV lithography,” Proceedings of theSPIE - The International Society for Optical Engineering, vol. 4561, pp. 255 –64, 2001.

[51] M. Lapisa, F. Zimmer, F. Niklaus, A. Gehner, and G. Stemme, “CMOS in-tegretable piston type micro mirror arrays for adaptive optics made of monocrystalline silicon using 3D integration,” in Proceedings IEEE InternationalConference on Micro Electro Mechanical Systems (MEMS), January 2009, pp.1007–1010.

www.silexmicrosystems.com

REFERENCES 69

[52] J. Sampsell, “Digital micromirror device and its application to projection dis-plays,” Journal of Vacuum Science & Technology B (Microelectronics andNanometer Structures), vol. 12, no. 6, pp. 3242 – 6, 1994.

[53] S. Kurth, R. Hahn, C. Kaufmann, K. Kehr, J. Mehner, U. Wollmann, W. Dotzel,and T. Gessner, “Silicon mirrors and micromirror arrays for spatial laser beammodulation,” Sensors and Actuators A (Physical), vol. A66, no. 1-3, pp. 76 –82, April 1998.

[54] F. Niklaus, S. Haasl, and G. Stemme, “Arrays of monocrystalline silicon mi-cromirrors fabricated using CMOS compatible transfer bonding,” Journal ofMicroelectromechanical Systems, vol. 12, no. 4, pp. 465 – 9, August 2003.

[55] T. Bakke, M. Friedrichs, B. Volker, M. Reiche, L. Leonardsson, H. Schenk, andH. Lakner, “Spatial light modulators with monocrystalline silicon micromirrorsmade by wafer bonding,” Proceedings of the SPIE - The International Societyfor Optical Engineering, vol. 5715, no. 1, pp. 69 – 79, 2004.

[56] T. Bakke, B. Volker, D. Rudloff, M. Friedrichs, H. Schenk, and H. Lakner,“Large scale, drift free monocrystalline silicon micromirror arrays made by waferbonding,” Proceedings of the SPIE - The International Society for Optical En-gineering, vol. 6114, pp. 611 402 – 1, January 2006.

[57] “International Technology Roadmap for Semiconductors (IRTS).” [Online].Available: www.irts.net

[58] M. Esashi, “Wafer level packaging of MEMS,” Transactions of the Institute ofElectronics, Information and Communication Engineers C, vol. J91-C, no. 11,pp. 527 – 33, November 2008.

[59] D. Van Thourhout, G. Roelkens, J. Van Campenhout, J. Brouckaert, andR. Baets, “Technologies for on-chip optical interconnects,” 2005 IEEE LEOSAnnual Meeting, pp. 204 – 5, 2005.

[60] M. Mastrangeli, S. Abbasi, C. Varel, C. Van Hoof, J. Celis, and K. Bohringer,“Self-assembly from milli-to nanoscales: methods and applications,” Journal ofMicromechanics and Microengineering, vol. 19, no. 8, pp. 083 001 (37 pp.) –,August 2009.

[61] I. Kurniawan, M. Tichem, and U. Staufer, “Dry self-alignment for batch-assembly of chips,” in Proceedings of IEEE Transducers, 2009, pp. 212–215.

[62] K.-F. Böhringer, K. Goldberg, M. Cohn, R. Howe, and A. Pisano, “Parallelmicroassembly with electrostatic force fields,” Proceedings IEEE InternationalConference on Robotics and Automation (Cat. No.98CH36146), vol. vol.2, pp.1204 – 11, 1998.

[63] C. Lin, F.-G. Tseng, and C.-C. Chieng, “Orientation-specific fluidic self-assembly process based on a capillary effect,” Journal of Micromechanics andMicroengineering, vol. 19, no. 11, pp. 115 020 (12 pp.) –, November 2009.

www.irts.net


[64] U. Srinivasan, M. Helmbrecht, C. Rembe, R. Muller, and R. Howe, “Fluidicself-assembly of micromirrors onto microactuators using capillary forces,” IEEEJournal of Selected Topics in Quantum Electronics, vol. 8, no. 1, pp. 4 – 11,January 2002.

[65] Y. A. Chapuis, A. Debray, L. Jalabert, and H. Fujita, “Alternative approachin 3D MEMS-IC integration using fluidic self-assembly techniques,” Journal ofMicromechanics and Microengineering, vol. 19, no. 2, pp. 105 002–105 011, 2009.

[66] R. Guerre et al., “Selective transfer technology for microdevice distribution,”Journal of Microelectromechanical Systems, vol. 17, no. 1, pp. 157 – 165, 2008.

[67] T. Grund, R. Guerre, M. Despont, and M. Kohl, “Transfer bonding technologyfor batch fabrication of SMA microactuators,” Eur. Phys. J. Special Topics, vol.158, pp. 237–242, 2008.

[68] M. Schmidt, “Wafer-to-wafer bonding for microstructure formation,” Proceed-ings of the IEEE, vol. 86, no. 8, pp. 1575 – 85, 1998.

[69] F. Niklaus, G. Stemme, J. Q. Lu, and R. J. Gutmann, “Adhesive wafer bonding,”Journal of Applied Physics, vol. 99, no. 3, pp. 031 101 1–28, February 2006.

[70] A. Goyal, J. Cheong, and S. Tadigadapa, “Tin-based solder bonding for MEMSfabrication and packaging applications,” Journal of Micromechanics and Micro-engineering, vol. 14, no. 6, pp. 819 – 25, July 2004.

[71] C.-W. Lin, H.-A. Yang, W. C. Wang, and W. Fang, “Implementation of three-dimensional SOI-MEMS wafer-level packaging using through-wafer interconnec-tions,” Journal of Micromechanics and Microengineering, vol. 17, no. 6, pp. 1200– 1205, 2007.

[72] P. Robogiannakis, E. Kyriakis-Bitzaros, K. Minoglou, S. Katsafouros,A. Kostopoulos, G. Konstantinidis, and G. Halkias, “Metallic bonding method-ology for heterogeneous integration of optoelectronic dies to CMOS circuits,”Microelectronic Engineering, vol. 85, no. 4, pp. 727 – 32, April 2008.

[73] R. Wolffenbuttel et al., “Low-temperature silicon wafer-to-wafer bonding usinggold at eutectic temperature,” Sensors and Actuators A (Physical), vol. A43,no. 1-3, pp. 223 – 9, 1994.

[74] A.-L. Tiensuu, M. Bexell, J.-A. Schweitz, L. Smith, and S. Johansson, “Assem-bling three-dimensional microstructures using gold-silicon eutectic bonding,”Sensors and Actuators A (Physical), vol. A45, no. 3, pp. 227 – 36, December1994.

[75] R. Wolffenbuttel, “Low-temperature intermediate Au-Si wafer bonding; eutecticor silicide bond,” Sensors and Actuators A (Physical), vol. A62, no. 1-3, pp. 680– 6, 1997.

REFERENCES 71

[76] S. Lani, A. Bosseboeuf, B. Belier, C. Clerc, C. Gousset, and J. Aubert, “Goldmetallizations for eutectic bonding of silicon wafers,” Microsystem Technologies,vol. 12, no. 10-11, pp. 1021 – 5, Sept. 2006.

[77] P. Chen, C. Lin, and C. Liu, “Amorphous Si/Au wafer bonding,” AppliedPhysics Letters, vol. 90, no. 13, pp. 132 120 – 1, March 2007.

[78] F. Niklaus, R. Kumar, J. McMahon, J. Yu, J.-Q. Lu, T. Cale, and R. Gut-mann, “Adhesive wafer bonding using partially cured benzocyclobutene forthree-dimensional integration,” Journal of the Electrochemical Society, vol. 153,no. 4, pp. 291 – 5, April 2006.

[79] F. Niklaus, A. Decharat, F. Forsberg, N. Roxhed, M. Lapisa, M. Populin,F. Zimmer, J. Lemm, and G. Stemme, “Wafer bonding with nano-imprint re-sists as sacrificial adhesive for fabrication of silicon-on-integrated-circuit (SOIC)wafers in 3D integration of MEMS and ICs,” Sensors and Actuators: A Physical,vol. 154, no. 1, pp. 180 – 6, August 2009.

[80] J. Oberhammer et al., “Selective wafer-level adhesive bonding with benzocy-clobutene for fabrication of cavities,” Sensors and Actuators A: Physical, vol.105, no. 3, pp. 297 – 304, 2003.

[81] Y. Feng, “Silicon direct bonding on vacuum and force,” IEEE InternationalConference on Mechatronics and Automation (IEEE Cat. No. 06EX1416C),pp. 491 – 6, 2006.

[82] S. H. Christiansen, R. Singh, and U. Gösele, “Wafer direct bonding: From ad-vanced substrate engineering to future applications in micro/nanoelectronics,”in Proceedings of the IEEE, vol. 94, no. 12, December 2006, pp. 2060–2106.

[83] P. Guegen, C. Ventosa, L. Di Cioccio, H. Moriceau, F. Grossi, M. Rivoire,P. Leduc, and L. Clavelier, “Physics of direct bonding: Applications to3D heterogeneous or monolithic integration,” Microelectronic Engineering,doi:10.1016/j.mee.2009.07.030, 2009.

[84] M. Chen, X. Yi, L. Yuan, and S. Liu, “Research on low-temperature an-odic bonding using induction heating,” Journal of Physics: Conference Series,vol. 34, no. 1, pp. 973 – 8, 2006.

[85] K. Knowles and A. T. J. van Helvoort, “Anodic bonding,” International Mate-rials Reviews, vol. 51, no. 5, pp. 273–311, October 2006.

[86] M. Taklo, P. Storas, K. Schjolberg-Henriksen, H. Hasting, and H. Jakobsen,“Strong, high-yield and low-temperature thermocompression silicon wafer-levelbonding with gold,” Journal of Micromechanics and Microengineering, vol. 14,no. 7, pp. 884 – 90, July 2004.

[87] G.-S. Park, Y.-K. Kim, K.-K. Paek, J.-S. Kim, J.-H. Lee, and B.-K. Ju, “Low-temperature silicon wafer-scale thermocompression bonding using electroplatedgold layers in hermetic packaging,” Electrochemical and Solid-State Letters,vol. 8, no. 12, pp. 330 – 2, December 2005.


[88] X. Ang, G. Zhang, B. Tan, J. Wei, Z. Chen, and C. Wong, “Direct metal tometal bonding for microsystems interconnections and integration,” Proceedingsof 7th Electronics Packaging Technology Conference, pp. 390–393, 2005.

[89] J. Kim, B. Jeong, M. Chiao, and L. Lin, “Ultrasonic bonding for MEMS sealingand packaging,” IEEE Transactions on Advanced Packaging, vol. 32, no. 2, pp.461–467, May 2009.

[90] R. Knechtel, “Glass frit bonding: An universal technology for wafer level en-capsulation and packaging,” Microsystem Technologies, vol. 12, no. 1-2 SPEC.ISS., pp. 63 – 68, 2005.

[91] T. Veenstra et al., “Use of selective anodic bonding to create micropump cham-bers with virtually no dead volume,” Journal of the Electrochemical Society, vol.148, no. 2, pp. 68 – 72, 2001.

[92] Y. Cheng et al., “Localized silicon fusion and eutectic bonding for MEMS fabri-cation and packaging,” Journal of Microelectromechanical Systems, vol. 9, no. 1,pp. 3 – 8, 2000.

[93] H.-A. Yang, M. Wu, and W. Fang, “Localized induction heating solder bondingfor wafer level MEMS packaging,” Journal of Micromechanics and Microengi-neering, vol. 15, no. 2, pp. 394 – 9, February 2005.

[94] J. Kim et al., “Thin layer laser bonding using spin-on-glass materials,” AppliedSurface Science, vol. 254, no. 21, pp. 6842 – 6848, 2008.

[95] W. van der Wijngaart, A. Ridgeway, and G. Stemme, “A micromachined knifegate valve for high-flow pressure regulation applications,” 12th InternationalConference on Solid-State Sensors, Actuators and Microsystems. Digest of Tech-nical Papers (Cat. No.03TH8664), vol. vol.2, pp. 1931 – 4, 2003.

[96] S. S. Nasiri et al., “Method and system of releasing a MEMS structure,” UnitedStates Patent 7250353, 2007.

[97] J. Bustillo, R. Howe, and R. Muller, “Surface micromachining for microelec-tromechanical systems,” Proceedings of the IEEE, vol. 86, no. 8, pp. 1552 – 74,August 1998.

[98] J. Buhler, F.-P. Steiner, and H. Baltes, “Silicon dioxide sacrificial layer etchingin surface micromachining,” Journal of Micromechanics and Microengineering,vol. 7, no. 1, pp. R1 – R13, 1997.

[99] C. W. Storment, D. A. Borkholder, V. Westerlind, J. W. Suh, N. I. Maluf,and G. T. Kovacs, “Flexible, dry-released process for aluminum electrostaticactuators,” Journal of Microelectromechanical Systems, vol. 3, no. 3, pp. 90 –96, 1994.

[100] K. Walsh, J. Norville, and Y.-C. Tai, “Photoresist as a sacrificial layer by dissolu-tion in acetone,” Technical Digest. MEMS 14th IEEE International Conferenceon Micro Electro Mechanical Systems (Cat. No.01CH37090), pp. 114 – 17, 2001.

REFERENCES 73

[101] D. Westberg, O. Paul, G. Andersson, and H. Baltes, “Surface micromachiningby sacrificial aluminium etching,” Journal of Micromechanics and Microengi-neering, vol. 6, no. 4, pp. 376 – 84, December 1996.

[102] O. Gigan et al., “Fabrication and characterization of resonant SOI micromechan-ical silicon sensors based on DRIE micromachining, freestanding release processand silicon direct bonding,” Proceedings of the SPIE - The International Societyfor Optical Engineering, vol. 4936, pp. 194 – 204, 2002.

[103] N. Somjit et al., “Novel concept of microwave MEMS reconfigurable 7x45◦ multi-stage dielectric-block phase shifters,” in Proceedings of MEMS, 2009.

[104] S. Metz et al., “Partial release and detachment of microfabricated metal andpolymer structures by anodic metal dissolution,” Journal of Microelectrome-chanical Systems, vol. 14, no. 2, pp. 383 – 391, 2005.

[105] S. Miyazaki and K. Nomura, “Development of perfect shape memory effectin sputter-deposited Ti-Ni thin films,” Proceedings of the IEEE Micro ElectroMechanical Systems, pp. 176 – 181, 1994.

[106] Y. Fu, H. Du, W. Huang, S. Zhang, and M. Hu, “TiNi-based thin films in MEMSapplications: A review,” Sensors and Actuators, A: Physical, vol. 112, no. 2-3,pp. 395 – 408, 2004.

[107] A. P. Lee, D. R. Ciarlo, P. A. Krulevitch, S. Lehew, J. Trevino, and M. A.Northrup, “A practical microgripper by fine alignment, eutectic bonding andSMA actuation,” Sensors and Actuators A: Physical, vol. 54, no. 1-3, pp. 755–759, June 1996.

[108] C. A. Ray, C. L. Sloan, A. D. Johnson, J. D. Busch, and B. R. Petty, “A siliconbased shape memory alloy microvalve,” in Mat. Res. Soc. Symp. Proc., vol. 276,1992, pp. 161–166.

[109] S. Miyazaki, M. Tomozawa, and H. Kim, “Development of high-speed microac-tuators utilizing sputter-deposited TiNi-base shape memory alloy thin films,”in Proceedings ACTUATOR, 2008, pp. 372–377.

[110] T. Mineta, K. Kasai, Y. Sasaki, E. Makino, T. Kawashima, and T. Shibata,“Flash-evaporated TiNiCu thick film for shape memory alloy micro actuator,”Microelectronic Engineering, vol. 86, pp. 1274–1277, 2009.

[111] K. Skrobanek, M. Kohl, and S. Miyazaki, “Stress-optimised shape memory mi-crovalves,” in Micro Electro Mechanical Systems, Proceedings, IEEE., TenthAnnual International Workshop on, 1997, pp. 256–261.

[112] P. J. Hesketh, J. S. Bintoro, and L. Luharuka, “Microvalve for fuel cells andminiature gas chromatographic system,” in Sensors Update 13, pp. 233–302.

[113] K. W. Oh and C. H. Ahn, “A review of microvalves,” Journal of Micromechanicsand Microengineering, vol. 16, no. 5, pp. R13 – R39, 2006.


[114] W. van der Wijngaart, A. Thorsén, and G. Stemme, “A study of orifice-controlled flow for microvalve design optimisation,” IEEE Journal of Micro-electromechanical Systems, vol. 14, pp. 200–206, April 2005.

[115] K. R. Williams, N. I. Maluf, E. N. Fuller, R. J. Barron, D. P. Jaeggi, and B. P.van Drieënhuizen, “A silicon microvalve for the proportional control of fluids,”in Proceedings of Transducers ’99, Sendai, Japan, 1999, pp. 1804–7.

[116] M. Walters, V. Dhuler, R. Mahadevan, A. Cowen, R. Wood, E. Hill, and I. Kao,“A silicon micromachined gate valve,” in 1998 Solid-State Sensors and ActuatorsSymposium, South Carolina, USA, 1998.

[117] A. P. Papavasiliou, D. Liepmann, and A. P. Pisano, “Fabrication of a free float-ing silicon gate valve,” in Proceedings of International Mechanical EngineeringCongress and Exposition (IMECE), Nashville, TN, USA, 1999.

[118] J. A. Frank and A. P. Pisano, “Low-leakage micro gate valves,” in Digest ofTechnical Papers, Transducers. 12th International Conference on Solid-StateSensors, Actuators and Microsystems, vol. 1, 2003, p. 143.

[119] R. Luharuka and P. Hesketh, “A bistable electromagnetically actuated rotarygate microvalve,” Journal of Micromechanics and Microengineering, vol. 18,no. 3, pp. 035 015 – 1, 2008.

[120] S. Booth and R. Kaine, “Fluid handling: Big gains from tiny valve,” ApplicanceDesign, pp. 46–48, April 2008. [Online]. Available: www.applicancedesign.com

[121] J. Uibel, “The miniaturization of flow control,” in 9th International Conferenceon fluid control, measurements, and visualization, 2007.

[122] M. Luckevich, “MEMS microvalves: the new valve world,” Valve-worldmagazine, pp. 79–83, May 2007. [Online]. Available: www.valve-world.net

[123] M. J. Zdeblick, R. Anderson, J. Jankoswki, B. Kline-Schode, L. Christel,R. Miles, and W. Weber, “Thermopneumatically actuated micro valves andintegrated electro-fluidic circuits,” in Technical Digest of Solid-state sensor andactuator workshop, Hilton Head Island, 1994, pp. 251–255.

[124] S. Messner, M. Muller, V. Burger, J. Schaible, H. Sandmaier, and R. Zengerle,“A normally-closed, bimetallically actuated 3-way microvalve for pneumatic ap-plications,” Proceedings MEMS. IEEE. Eleventh Annual International Work-shop on Micro Electro Mechanical Systems. An Investigation of Micro Struc-tures, Sensors, Actuators, Machines and Systems (Cat. No.98CH36176), pp. 40– 44, 1998.

[125] H. Jerman, “Electrically activated normally closed diaphragm valves,” Journalof Micromechanics and Microengineering, vol. 4, no. 4, pp. 210 – 216, December1994.

www.applicancedesign.com

www.valve-world.net

REFERENCES 75

[126] C. Fu, Z. Rummler, and W. Schomburg, “Magnetically driven micro ball valvesfabricated by multilayer adhesive film bonding,” Journal of Micromechanics andMicroengineering, vol. 13, no. 4, pp. 96 – 102, July 2003.

[127] X. Yang, A. Holke, S. Jacobson, J. Lang, M. Schmidt, and S. Umans, “Anelectrostatic, on/off microvalve designed for gas fuel delivery for the MIT mi-croengine,” Journal of Microelectromechanical Systems, vol. 13, no. 4, pp. 660– 668, August 2004.

[128] H. Li, D. Roberts, J. Steyn, K. Turner, O. Yaglioglu, N. Hagood, S. Spearing,and M. Schmidt, “Fabrication of a high frequency piezoelectric microvalve,”Sensors and Actuators A (Physical), vol. A111, no. 1, pp. 51 – 56, March 2004.

[129] E.-H. Yang, C. Lee, J. Mueller, and T. George, “Leak-tight piezoelectric mi-crovalve for high-pressure gas micropropulsion,” Journal of Microelectromechan-ical Systems, vol. 13, no. 5, pp. 799 – 807, October 2004.

[130] P. Shao, Z. Rummler, and W. Schomburg, “Polymer micro piezo valve with asmall dead volume,” Journal of Micromechanics and Microengineering, vol. 14,no. 2, pp. 305 – 9, February 2004.

[131] M. A. Huff, J. R. Gilbert, and M. A. Scmidt, “Flow characteristics of a pressurebalanced microvalve,” in Proceeding IEEE Transducers, June 1993, pp. 98–101.

[132] P. W. Barth, C. C. Beatty, L. A. Field, J. W. Baker, and G. B. Gordon, “Arobust normally-closed silicon microvalve,” in Technical Digest of Solid-statesensor and actuator workshop, Hilton Head Island, 1994, pp. 248–250.

[133] X. Yang, C. Grosjean, and Y.-C. Tai, “Design, fabrication, and testing of micro-machined silicone rubber membrane valves,” Journal of MicroelectromechanicalSystems, vol. 8, no. 4, pp. 393 – 402, December 1999.

[134] D. Bosch, B. Heimhofer, G. Muck, H. Seidel, U. Thumser, and W. Welser,“A silicon microvalve with combined electromagnetic/electrostatic actuation,”Sensors and Actuators A (Physical), vol. A37-A38, pp. 684 –692, June 1993.

[135] M. Kohl, K. Skrobanek, and S. Miyazaki, “Development of stress-optimisedshape memory microvalves,” Sensors and Actuators A (Physical), vol. A72,no. 3, pp. 243 – 50, February 1999.

[136] M. Kohl and B. Krevet, “3D simulation of a shape memory microactuator,”Material Transactions, vol. 43, no. 5, pp. 1030–1036, 2002.

[137] M. Kohl, D. Dittmann, E. Quandt, and B. Winzek, “Thin film shape memorymicrovalves with adjustable operation temperature,” Sensors and Actuators A(Physical), vol. A83, no. 1-3, pp. 214 – 219, May 2000.

[138] M. Kohl, D. Dittmann, E. Quandt, B. Winzek, S. Miyazaki, and D. Allen,“Shape memory microvalves based on thin films or rolled sheets,” MaterialsScience and Engineering A, vol. 273, pp. 784 – 788, 1999.


[139] M. Kohl, J. Gobes, and B. Krevet, “Normally-closed shape memory microvalve,”International Journal of Applied Electromagnetics and Mechanics, vol. 12, no.1-2, pp. 71 – 77, 2000.

[140] C. Clos, “A study of non-blocking switching networks,” Bell System technicaljournal, vol. 32, pp. 406–424, March 1953.

[141] D. K. Hunter, “Switching systems,” Encyclopedia of Information Technology,vol. 42, pp. 335–370, 2000.

[142] J. S. Turner and R. Melen, “Multirate Clos networks,” IEEE CommunicationsMagazine, vol. 41, no. 10, pp. 38–44, October 2003.

[143] [Online]. Available: http://en.wikipedia.org/wiki/Crossbar_switch

[144] [Online]. Available: http://en.wikipedia.org/wiki/Nonblocking_minimal_spanning_switch

[145] T. Kanai, S. Umemura, S. Inagaki, and S. Hosokawa, “High density pin boardmatrix switches for automated MDF systems,” IEEE Transactions on Compo-nents, Packaging, and Manufacturing Technology, vol. 15, no. 5, pp. 893–903,October 1992.

[146] S. Inagaki, T. Kanai, and S. Hosokawa, “High-density pin-board matrix switchand its crosstalk characteristics,” Electronics and Communications in Japan,vol. 78, no. 7, pp. 42–52, July 1995.

[147] H. Isobe, M. Ito, and T. Murakami, “Automatic MDF (main distribution frame)apparatus,” US Patent 6295483, September 2001.

[148] H. Hosaka, H. Kuwano, and K. Yanagisawa, “Electromagnetic microrelays: Con-cepts and fundamental characteristics,” Sensors and Actuators, A: Physical,vol. 40, no. 1, pp. 41 – 47, 1994.

[149] H. Hosaka, H. Kuwano, and K. Murata, “Development of miniature matrix re-lays for telecommunication systems,” Advances in Information Storage Systems,vol. 10, pp. 275–285, 1998.

[150] G. M. Rebeiz and R. Muldavin, “RF MEMS switches and switch circuits,” IEEEMicrowave Magazine, vol. 2, no. 4, pp. 59–71, December 2001.

[151] H. A. C. Tilmans, “MEMS components for wireless communications (invitedpaper),” in Proceedings EUROSENSORS XVI, September 2002, pp. 1–34.

[152] G. Rebeiz, “RF MEMS switches: status of the technology,” in TRANSDUCERS,Solid-State Sensors, Actuators and Microsystems, 12th International Confer-ence on, vol. 2, 2003, pp. 1726–1729.

[153] P. Farinelli, F. Giacomozzi, G. Mannocchi, R. Marcelli, B. Margesin, P. Mez-zanotte, S. Di Nardo, P. Russer, R. Sorrentino, F. Vitulli, and L. Vietzorreck,“RF-MEMS SPDT switch on silicon substrate for space applications,” in Topi-cal Meeting on Silicon Monolithic Integrated Circuits in RF Systems, Digest ofPapers., 2004, pp. 151–154.

http://en.wikipedia.org/wiki/Crossbar_switch

http://en.wikipedia.org/wiki/Nonblocking_minimal_spanning_switch

REFERENCES 77

[154] B. Yassini, S. Choi, A. Zybura, M. Yu, R. Mihailovich, and J. DeNatale, “A novelMEMS LTCC switch matrix,” in Microwave Symposium Digest,IEEE MTT-SInternational, vol. 2, 2004, pp. 721–724.

[155] E. McErlean, J.-S. Hong, S. Tan, L. Wang, Z. Cui, R. Greed, and D. Voyce,“2x2 RF MEMS switch matrix,” IEEE Proceedings on Microwaves, Antennasand Propagation, vol. 152, no. 6, pp. 449–454, December 2005.

[156] S. DiNardo, P. Farinelli, G. F., G. Mannochi, R. Marcelli, B. Margesin, P. Mez-zanotte, V. Mulloni, P. Russer, R. Sorrentino, F. Vitulli, and L. Vietzorreck,“RF MEMS based switch matrices for complex switching networks,” in Proc. ofMicrowave Technology and Techniques Workshop, ESA ESTEC, May 2005.

[157] R. Mansour, “RF MEMS for space applications,” in Proceedings of InternationalConference on MEMS, NANO and Smart Systems, 2005, pp. 191–192.

[158] M. Daneshmand and R. Mansour, “Monolithic RF MEMS switch matrix inte-gration,” in Microwave Symposium Digest, IEEE MTT-S International, 2006,pp. 140–143.

[159] S. Patton and J. Zabinski, “Fundamental studies of Au contacts in MEMS RFswitches,” Tribology Letters, vol. 18, no. 2, pp. 215 – 230, February 2005.

[160] H. Kwon, D.-J. Choi, J.-H. Park, H.-C. Lee, Y.-H. Park, Y.-D. Kim, H.-J. Nam,Y.-C. Joo, and J.-U. Bu, “Contact materials and reliability for high power RF-MEMS switches,” in Proceedings of the IEEE International Conference on MicroElectro Mechanical Systems (MEMS), 2007, pp. 231 – 234.

[161] L. Chen, H. Lee, Z. Guo, N. McGruer, K. Gilbert, S. Mall, K. Leedy, andG. Adams, “Contact resistance study of noble metals and alloy films using ascanning probe microscope test station,” Journal of Applied Physics, vol. 102,no. 7, pp. 074 910 – 1, October 2007.

[162] H. Kwon, S.-S. Jang, Y.-H. Park, T.-S. Kim, Y.-D. Kim, H.-J. Nam, and Y.-C. Joo, “Investigation of the electrical contact behaviors in Au-to-Au thin-filmcontacts for RF MEMS switches,” Journal of Micromechanics and Microengi-neering, vol. 18, no. 10, p. 105010 (9p), October 2008.

[163] V. Jaecklin, C. Linder, N. de Rooij, J.-M. Moret, and R. Vuilleumier, “Line-addressable torsional micromirrors for light modulator arrays,” Sensors andActuators A (Physical), vol. A41, no. 1-3, pp. 324 – 9, 1994.

[164] D. A. Winick, B. E. Duewer, S. Palchaudhury, and P. D. Franzon, “Performanceevaluation of micromechanical binary phase-only holographic optical elements,”in 47th Electronic Components and Technology Conference, May 1997, pp. 419–424.

[165] D. A. Winick, B. E. Duewer, S. Chaudhury, J. M. Wilson, J. Tucker, U. Eksi,and P. D. Franzon, “MEMS-based diffractive optical-beam-steering technology,”in Proc. SPIE - Int. Soc. Opt. Eng. (USA), vol. 3276, no. 1, 1998, pp. 81–87.


[166] B. Cunningham, J. Bernstein, D. Seltzer, and D. Hom, “Micromirror pixel ad-dressing using electromechanical bistability,” in Technical Digest. Solid-StateSensor and Actuator Workshop, 1998, pp. 285 – 7.

[167] G. Zhou, L. VJ, F. S. Chau, and F. E. H. Tay, “Line-addressable digital-deflection programmable micromirror array,” Opt. Lett., vol. 29, no. 22, pp.2581–2583, 2004.

[168] B. E. Duewer, J. M. Wilson, D. A. Winick, and P. D. Franzon, “ProgrammableMEMS capacitor arrays,” in Proceedings of the SPIE, vol. 3893, no. 1, 1999, pp.262–271.

[169] A. Jourdain, P. De Moor, K. Baert, I. De Wolf, and H. Tilmans, “Mechanicaland electrical characterization of BCB as a bond and seal material for cavitieshousing (RF-)MEMS devices,” Journal of Micromechanics and Microengineer-ing, vol. 15, no. 7, pp. 89 – 96, July 2005.

[170] J. Thurn and R. Cook, “Stress hysteresis during thermal cycling of plasma-enhanced chemical vapor deposited silicon oxide films,” Journal of AppliedPhysics, vol. 91, no. 4, pp. 1988 – 92, February 2002.

[171] J. Thurn, R. Cook, M. Kamarajugadda, S. Bozeman, and L. Stearns, “Stresshysteresis and mechanical properties of plasma-enhanced chemical vapor de-posited dielectric films,” Journal of Applied Physics, vol. 95, no. 3, pp. 967 –76, February 2004.

[172] Z. Cao and X. Zhang, “Mechanism of temperature-induced plastic deformationof amorphous dielectric films for MEMS applications,” in Proceedings of the 18thIEEE International Conference on Micro Electro Mechanical Systems (MEMS),2005, pp. 471–474.

Wafer-level heterogeneous integration of MEMS actuators

Documents

Transcript of Wafer-level heterogeneous integration of MEMS actuators