Assembly of microsystems for optical and fluidic applications

(1)

Assembly of Microsystems for Optical and Fluidic

Applications

Sjoerd Haasl

MICROSYSTEM TECHNOLOGY

DEPARTMENT OF SIGNALS, SENSORS AND SYSTEMS ROYAL INSTITUTE OF TECHNOLOGY

ISBN 91-7283-958-9 ISSN 0281-2878 TRITA-ILA-0501

Submitted to the School of

Computer Science — Electrical Engineering — Engineering Physics (DEF), Royal Institute of Technology (KTH), Stockholm, Sweden,

in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy Stockholm 2005

(2)

Copyright c2005 by Sjoerd Haasl

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

The copyrights for the appended journal papers belong to the publishing houses of the journals concerned. The copyrights for the appended manuscripts belong to their authors.

Printed by Universitetsservice US AB, Stockholm 2005.

Thesis for the degree of Doctor of Philosophy at the Royal Institute of Technology, Stockholm, Sweden, 2005.

(3)

ABSTRACT iii

Abstract

This thesis addresses assembly issues encountered in optical and ﬂuidic microsystem applications.

In optics, the ﬁrst subject concerns the active alignment of components in optical ﬁber systems. A solution for reducing the cost of optical component assembly while retaining submicron accuracy is to integrate the alignment mechanism onto the optical substrate. A polymer V-shaped actuator is presented that can carry the weight of the large components – on a micromechanical scale – and that can generate movement with six degrees of freedom.

The second subject in optics is the CMOS-compatible fabrication of monocrystalline silicon micromirror arrays that are intended to serve as CMOS-controlled high- quality spatial light modulators in maskless microlithography systems. A wafer-level assembly method is presented that is based on adhesive wafer bonding whereby a monocrystalline layer is transferred onto a substrate wafer in a CMOS-compatible process without needing bond alignment.

In fluidics, a hybrid assembly method is introduced that combines two separately micromachined structures to create hotwire anemometers that protrude from a surface with minimum interference with the air flow. The assembled sensor enables one to make accurate time-resolved measurements of the wall shear stress, a quantity that has previously been hard to measure with high time resolution. Also in the field of hotwire anemometers, a method using a hotwire anemometer array is presented for measuring the mass flow, temperature and composition of a gas in a duct.

In biochemistry, a bio-analysis chip is presented. Single nucleotide polymorphism scoring is performed using dynamic allele-speciﬁc hybridization (DASH). Using monolayers of beads, multiplexing based on single-bead analysis is achieved at heating rates more than 20 times faster than conventional DASH provides.

Space and material efficiency in packaging are the focus of the other two projects in fluidics. The first introduces an assembly based on layering conductive adhesives for the fabrication of miniature polymer electrolyte membrane fuel cells. The fuel cells made with this low-cost approach perform among the best of their type to date.

The second project concerns a new cross-ﬂow microvalve concept. Intended as a step towards the mass production of large-ﬂow I/P converters, the silicon footprint area is minimized by an out-of-plane moving gate and in-plane, half-open pneumatic chan- nels.

Keywords: microsystem technology, micromachining, assembly, active alignment, BCB, polymer, V-shaped actuator, micromirror, CMOS-compatible, membrane transfer bonding, CTA, hotwire, wall shear stress, gas-composition measurement, single nucleotide polymorphism, SNP scoring, dynamic allele-speciﬁc hybridization, DASH, miniature fuel cell, conductive adhesives, I/P converter, cross-ﬂow microvalve

Sjoerd Haasl

Microsystem Technology, Department of Signals, Sensors and Systems Royal Institute of Technology, SE-100 44 Stockholm, Sweden

(4)

(5)

ERRATUM v

Erratum

Paper 2, page 242, paragraph 4, line 3, states: “For the present MEMS sensors, the wire and supports used are made of 2-µm thick aluminium layers, and though the supports are additionally strengthened by a material that acts as a thermal insulator this will be neglected in the heat rate calculations.”

This should read: “For the present MEMS sensors, the wire and support leads used are made of 2-µm thick aluminium layers, and although the supports are additionally strengthened by a material that acts as a thermal conductor this will be neglected in the heat rate calculations to provide a worst-case analysis.”

(6)

(7)

vii

To my Elin

(8)

(9)

CONTENTS ix

List of papers

The presented thesis is based on the following journal papers:

1. Hybrid mounted micromachined aluminum hotwires for wall shear stress mea- surements

Sjoerd Haasl, Dirk Mucha, Valery Chernoray, Thorbj¨orn Ebefors, Peter Enoks- son, Lennart L¨ofdahl and G¨oran Stemme

Journal of Microelectromechanical Systems, accepted for publication.

2. Characteristics of a hot wire microsensor for time-dependent wall shear stress measurements

Lennart L¨ofdahl, Valery Chernoray, Sjoerd Haasl, G¨oran Stemme and Mihir Sen

Experiments in Fluids, vol. 35, pp. 240–51, 2003.

3. Implementation of a hotwire array sensor for ﬂuid property analysis via ﬂow velocity distribution measurements

Mattias Wind˚a, Sjoerd Haasl, Patrik Melv˚as and G¨oran Stemme Submitted for journal publication.

4. Arrays of monocrystalline silicon micromirrors fabricated using CMOS compat- ible transfer bonding

Frank Niklaus, Sjoerd Haasl and G¨oran Stemme

Journal of Microelectromechanical Systems, vol. 12, no. 4, pp. 465–9, 2003.

5. Adhesive copper ﬁlms for an air-breathing polymer electrolyte fuel cell

Fréd´eric Jaouen, Sjoerd Haasl, Wouter van der Wijngaart, Anders Lundblad, Göran Lindbergh and Göran Stemme

Journal of Power Sources, accepted for publication.

6. Robust, large-deﬂection, in-plane thermal polymer V-shaped actuators

Sjoerd Haasl, Patrick Griss, Thorbj¨orn Ebefors, Hans Sohlström, Edvard Kälvesten and Göran Stemme, manuscript for journal publication.

7. Genotyping by dynamic heating of monolayered beads on a microheated surface Aman Russom, Sjoerd Haasl, Anna Ohlander, Torsten Mayr, Anthony J.

Brookes, Helene Andersson and G¨oran Stemme Electrophoresis, vol. 25, no. 21–22, pp. 3712–9, 2004.

8. Out-of-plane knife-gate microvalves for controlling large gas ﬂows

Sjoerd Haasl, Stefan Braun, Samir Sadoon, Wouter van der Wijngaart and G¨oran Stemme

Submitted for journal publication.

(12)

The contribution of Sjoerd Haasl to the publications listed above is:

1 Major part of design, fabrication and writing, part of experiments 2 Major part of design and fabrication, part of experiments and writing 3 Part of design and writing

4 Part of design, experiments and writing 5 Part of design, experiments and writing

6 Major part of design, all fabrication, experiments and writing

7 Major part of design and fabrication, part of experiments and writing 8 Part of fabrication, experiments and writing

The work has been presented at the following international conferences:

1. Hybrid mounted micromachined aluminum hotwire for near-wall turbulence mea- surements

Sjoerd Haasl, Dirk Mucha, Valery Chernoray, Thorbj¨orn Ebefors, Peter Enoks- son, Lennart L¨ofdahl and G¨oran Stemme

Proceedings MEMS 2002, Las Vegas, USA, Jan., 2002, pp. 336–9, poster pre- sentation.

2. Arrays of monocrystalline silicon micromirrors fabricated using CMOS compat- ible transfer bonding

Sjoerd Haasl, Frank Niklaus and G¨oran Stemme

Proceedings MEMS 2003, Kyoto, Japan, Jan., 2003, pp. 271–4, poster presenta- tion.

3. Measurement of the turbulence intensities in a ﬂat plate boundary layer G. Michell, Valery Chernoray, Lennart L¨ofdahl, Sjoerd Haasl and G¨oran Stemme

4th symposium on Turbulence, Heat and Mass Transfer, Antalya, Turkey, Oct., 2003, oral presentation.

4. Robust, large-deﬂection, in-plane thermal polymer V-shaped actuators

Sjoerd Haasl, Patrick Griss, Hans Sohlstr¨om, Edvard K¨alvesten and G¨oran Stemme

Proceedings MEMS 2004, Maastricht, The Netherlands, Jan. 25–29, 2004, pp. 510–

3, poster presentation.

5. Time-resolved wall shear stress measurements using MEMS

Alister N. Gibson, Valery Chernoray, Lennart L¨ofdahl, Sjoerd Haasl and G¨oran Stemme

XXI International Congress of Theoretical and Applied Mechanics, Warsaw, Poland, Aug., 2004, oral presentation.

(13)

LIST OF PAPERS xiii

6. Genotyping by dynamic heating of monolayered beads on a microheated surface Aman Russom, Sjoerd Haasl, Anna Ohlander, Helene Andersson and G¨oran Stemme

Micro Total Analysis Systems, Malm¨o, Sweden, Sept., 2004, poster presentation.

(14)

(15)

1 INTRODUCTION 1

1 Introduction

1.1 Microsystems and assembly

Assembly is defined as the “fitting together of manufactured parts into a complete machine, structure or unit of a machine” [1]. This can occur anywhere in the process of making a micro-electromechanical system (MEMS), most commonly towards the end of a process but also at intermediate stages. Because of the individuality of assembly, it is inherently more expensive than batched microelectronic processes, so it always has to be economically justifiable. Assembly is about interfacing. Assembly is the first step when different domains in engineering are combined, the next step being integration. As an illustration of the multidisciplinary nature of MEMS, the following table gives an overview of the engineering fields interfaced in the work presented in this thesis.

Table 1: A classiﬁcation of the presented work according to the ﬁelds of engineering to which they belong.

Field: Electrical Optical Fluidic Mechanical Chemical

Micromirrors × × ×

Alignment × × ×

DASH × × × ×

Anemometer × ×

Fuel cell × × ×

Microvalve × × ×

1.2 Reasons for assembly

Ultimately, both in research and in production, achieving the goal (in its broadest sense) at minimum cost is the only objective. This cost can be measured both in time and money. In research, more frequently than in production, time is cheaper than money, and it is in that spirit that the devices in Chapters 4 and5 have been designed. Most often, however, within applied research ﬁelds such as MEMS, the target is production, where time is money, so both need to be minimized (Chapters2, 3and6).

Sometimes assembly can solve problems that, with the current state of the art, cannot be solved in any other way (Chapters 2, 3). This reverts in a sense to the original statement, since reaching a goal at any reasonable cost is better than not reaching it at all. In some of the papers accompanying this thesis, the aim is to reduce the existing assembly costs through intelligent design or added functionality to the point where mass production becomes feasible (Chapters2,6 and7).

(16)

1.3 Objective

The objectives of this thesis are to describe the role of assembly in MEMS and to present some solutions to the problems that arise within the ﬁeld, and both in the perspective of their alternatives. It is important to note that, in all of the applications, assembly is the means and never the end.

1.4 Structure

This thesis is a compilation of the results of six freestanding projects. They are divided in two categories, based on the subﬁeld of MEMS to which they belong:

optical and ﬂuidic MEMS. The chapters give background information on the subject starting from a general engineering standpoint, highlight the application of assembly, and provide a discussion that, in some cases, goes beyond the published results. All the results are presented in the paper reprints and manuscripts in the ﬁnal section of this thesis.

(17)

2 ACTIVE ALIGNMENT IN TELECOMMUNICATIONS TECHNOLOGY 3

2 Active alignment in telecommunications technology

2.1 Introduction

One of the most expensive steps and hence one of the major obstacles for the mass production of photonic components is the assembly. Photons tend to follow straight paths unless reﬂected or refracted, which means that alignment to six degrees of freedom (three axial and three rotational) is essential to control their ﬂow and loss.

Every connection within an optical assembly must have controlled alignment up to the nanometer level to minimize the photon loss as the optical signal passes through [2].

The solutions to the alignment problem can involve passive alignment or active alignment. Passive alignment uses the structure of the optical components and the substrate they are placed on, the state-of-the-art in passive alignment is capable of achieving a precision of 0.5 µm [3,4]. Although this is suﬃcient in many cases and by far the cheapest alternative, a considerable number of high-end applications need more precision. This is where active alignment comes into the picture. Active alignment uses a feedback system while aligning the diﬀerent components to the substrate.

This system may involve monitoring the loss through the system during alignment or using a microscope to observe the alignment. For this, alignment machines with a price tag of typically USD 400,000 [5] are needed for large-scale production. Often a combination of both active and passive alignment methods is used.

One way to reduce assembly costs in the long run is to integrate an active alignment system into the optical substrate. This would reduce the degree of accuracy needed for the positioning system that places the components on the substrate. An integrated system that bridges the last 10 µm of inaccuracy to an accuracy of 0.1 µm would be a large step towards fully integrated active alignment. As yet, no satisfactory solutions are available on the market.

An important element to consider when designing non-integrated on-chip active alignment structures is that the actuators have to be robust enough to survive the placement of the relatively bulky optical components. Although there is no typical laser chip size, at least two of the three dimensions usually exceed 500 µm, and the same is true for the lenses used to guide the light into the fiber. The fiber itself typi- cally has a diameter of 125 µm but the force required to manipulate it increases with decreasing free-hanging length and thus package size. The thin structures produced by surface micromachining are often not sufficient to carry these loads, which makes bulk micromachining the preferred fabrication method. The choice of materials is limited to non-outgassing materials to avoid interference with the free-space optics.

2.2 On-chip active alignment methods

Although theoretically, on-chip active alignment could be achieved by adapting the properties of the components to the position in which they are ﬁxed, for example, by changing the lens’s focal length [6], it is achieved most often by physically moving the components. Table2 lists a selection of the micromachined active optical alignment methods that can be found in literature.

(18)

SjoerdHaasl:AssemblyofMicrosystemsforOpticalandFluidicApplications Table 2: An overview of micromachined active optical alignment principles.

Object Principle Method DOF Stroke Fabr. Reference

Fiber E-static Metallized ﬁber 2 T XY 25µm Bulk Hockaday [7], Jebens

et al. [8], Kikuya et al. [9,10]

Fiber E-magnetic Electroplated FeNi 1 T 150µm N/A Nagaoka [11]

Fiber E-magnetic Permanent magnet 2 T XY 100µm Bulk Gerlach et al. [12]

Fiber Thermal Arch beam (XZ) & 3 T XZ 30µm, Y 120 µm Bulk Wood et al. [13,14]

Bimorph (Y)

Fiber Thermal Folded buckling mode 1 T X 12µm Bulk Syms et al. [15]

Fiber Thermal Thick arm/thin arm 1 T X 125µm Bulk Field et al. [16]

Fiber Thermal Selective current 1 T X 400µm Bulk Kopka et al. [17]

Fiber SMA Bending arc 1 T Y 125µm N/A Jebens et al. [8]

Fiber Piezoelectric Mono- and bimorph 2 T XY 100µm Bulk Gerlach et al. [12]

cantilever

Lens (int.) E-static Comb drive 2 T XZ 48µm Bulk Kim et al. [18]

Lens (int.) E-static Scratch drive 3 T XZ 110µm, Y 250 µm Surface Fan et al. [19]

Lens (ext.) E-static Scratch drive 2 T (1 R) XZ 110µm, Y 250 µm Surface Fan et al. [20]

Lens (ext.) E-static Comb drive 1 T X 72µm Bulk Grade et al. [21]

Mirrors & lens E-static Scratch drive 2 R, 1 T ∼90^◦, 80µm Surface Lin et al. [22]

Mirrors Thermal Hot arm/cold arm 2 R (2 T) 2.5^◦ Surface Ishikawa et al. [23]

Mirrors E-static Vibromotor 2 R, 1 T ∼2^◦, 80µm Surface Solgaard et al. [24,25]

Mirrors E-static Parallel plate 1 R 1^◦ Surface Zhang et al. [26]

Platform Thermal Buckling mode 1 T X 110µm Bulk Syms et al. [27]

Platform Thermal V-shaped actuator 1 T X 40µm Bulk This work

Platform Thermal V-shaped actuator 1 T Y 9µm Bulk This work

(19)

Notes for Table2:

Object: The object that is moved. A diﬀerence is made between integrated and external lenses. For the mirror- and lens-based solutions, references are given only where optical alignment was speciﬁed as an application.

DOF: Degrees of freedom can be translational (T) or rotational (R).

DOF in parenthesis indicates the possible degrees of freedom that were not used in the referred publication.

Stroke: Y is the out-of-plane axis; Z is the in-plane axis in the direction of the optical beam (if applicable).

For alignment solutions where the fiber was moved, the stroke of the loaded actuator is given, i.e. where the fiber stiffness counteracts the actuator movement. The exception is Wood et al. [13] where only the unloaded stroke is reported.

For mirrors, the maximum static deﬂection is given, not the resonant.

Fabrication: This column indicates whether the device is fabricated using surface or bulk micromachining.

When moving fibers, bulk micromachining is used since the stiffness of the fiber needs to be overcome. A straightforward way of manipulating the fiber is to electrostatically actuate a fiber coated with a conductive layer in a V-shaped groove where the walls are coated with a metal [9,10,7,8]. Related to this method, the coating of the fiber with an FeNi alloy [11] or the attachment of a permanent magnet [12] enables electromagnetic fields to be used to actuate the fiber. Another way is to combine piezoelectric monomorphs and bimorphs to form a 2-dimensional fiber aligner [12].

The solution that has come the farthest in full optical fiber alignment is Boeing’s in- package MEMS aligner [13,14], which combines two in-plane thermal shape actuators with an out-of-plane thermal bimorph actuator. Also based on thermal actuation, several single-dimension fiber alignment methods have been devised [15,16,17]. In effect all hot-arm/cold-arm actuators, they differ in the way of heating and in how the arms are coupled to generate a linear movement. Finally, the method using a shape-memory alloy (SMA) wire arc should be noted [8]. The wire arc is clamped against the fiber end and moves it when current is passed through.

For the components that can be moved in their entirety, i.e., lenses and mirrors, electrostatic actuation principles are most common. Parallel-plate actuators [26], which use the electrostatic attraction between two parallel plates, and comb-drive actuators [18,21], a variant where the capacitor plates are interdigitated to create a constant force over a larger displacement, are well-known and easily controlled actuation principles that consume little power. Electrostatic vibromotors [24,25], which push a slider forward with vibrating comb-drive actuated pistons impacting at an angle against both sides, and scratch-drive actuators [28,19,20], where the center part of a plate leaning on its bushing is repeatedly pulled down electrostatically, causing it to inch forward, can move objects over large distances with high resolution. Surface micromachined hot-arm/cold-arm thermal actuators, which are based on heating up one leg of a U-shaped actuator more than the other to generate a rotation, have also been used to move mirrors with four degrees of freedom.

A more general approach is to use a platform that can carry even the larger optical components. One device uses two sets of thermal buckling-mode-beam actuators [27].

(20)

These sets consist of slightly pre-buckled beams that expand transversely upon heating and are placed in such a way that the platform is very rigid for in-plane rotation and lateral motion. The second platform approach is the polymer-based V-shaped actuator described below.

2.3 BCB

BCB is a collective name for benzocyclobutene-based polymers. It is produced by Dow Chemical under the trade name Cyclotene and was developed for the microelectronics industry. Its thermal stability, low curing temperature, good chemical resistance and compatibility with various metallization systems enables it to be used in a wide range of applications for planarization [29], mechanical support and protection [30,31]. Its low dielectric constant and its low loss at high frequencies make it suitable as a structural material in RF-MEMS components [32,33] and high density printed-circuit- board technology [34]. BCB is also used as an optical material for wave guides (both core and cladding). That BCB does not outgas after precuring at 70 ^◦C makes it a good material for adhesive bonding, in which outgassing can aﬀect the bonding quality [35]. BCB is usually spun on the wafer but is also available as laminate (Sumotomo Bakelite APL-4901, [34]). It is available both in photosensitive and dry- etch grade. One property of polymers that is often seen as a problem is their thermal coeﬃcient of expansion (CTE). With a CTE of about 50 ppm/^◦C, BCB is considered to be a low-CTE polymer, even though its CTE still is 10 times greater than that of silicon.

In this work, BCB was chosen as an alternative to the polyimide (PI) that was used in the previously reported V-groove actuators [36,37] because it has a similar CTE as well as a good track record for optical applications. Furthermore, it has less static shrinkage than polyimide, a feature that enhances the controllability of the fabrication. Young’s moduli for both BCB and polyimide lie between 2 and 4 GPa.

The following table compares the relevant parameters.

Table 3: Some properties of BCB and polyimide relevant to actuation at room tempera- ture [38,37,39,40].

Young’s modulus Thermal conductivity Heat capacity (C)

(GPa) (W/m·K) (J/kg·K)

BCB 4.2 0.29 2100

Polyimide 2-3 0.16-0.25 2000

The BCB used in this work was dry-etch BCB. It was selected over photosensitive BCB for ﬁnancial reasons: photosensitive BCB is 4 times more expensive than the dry-etch variant, and, because it has a 1-week shelf life at room temperature while the shelf life of dry-etch BCB is 1 year, it is not useful for research purposes, for which small amounts need to be thawed out and unused portions thrown away. Dry-etch BCB, however, still presented some problems because of the residue after the dry- etching process. This was resolved by using the residue lift-oﬀ process described in Paper 6.

(21)

2.4 Polymer V-shaped actuators

To solve the problem of supporting and moving the relatively large and heavy optical components like lasers and lenses, an actuator was designed based on the polyimide V-groove actuators earlier developed at the department [36,41]. The principle of these actuators (shown in Figure1) is the rotary movement caused by the larger absolute thermal expansion of the polymer at the wide end of the V groove and than at the narrow end. A lever transforms this rotary movement to a near-linear movement in one dimension. Figure 2 illustrates how the principle can be implemented both in plane and out of plane.

In previous work on the V-groove actuators [36,41,42], explicit use was made of the larger static shrinkage of polyimide when it is cured after the release etch instead of before it. Even though static shrinkage is acceptable when creating actuators for optical alignment, designing a platform is easier if this shrinkage is reduced to a minimum. Therefore, curing was performed before the free etching of the actuators.

In the out-of-plane V-groove actuators, the actuator angle γ (see Figure 1) is 2× 54.74^◦, which is defined by the KOH etch that creates it. Given the increased design freedom when designing in plane, several different actuator variants were designed. Figure3shows five variants that were fabricated. Variant X1 (a) shows a pure V-shaped actuator, similar to the out-of-plane actuator. In variant X2 (b), hook-like BCB-embedded silicon prongs were added to provide more efficient heating and increased robustness. Since the BCB in this variant envelops the silicon prongs, the actuator is no longer as dependent on the adhesion between the silicon and the BCB, as it is in Variant X1. No problems occurred during measurement, but it was the silicon-BCB interface that failed during manual destruction of a Variant X1 device.

In Variant X3 (Figure 3c), polysilicon leads were placed over a silicon joint to enable electrical heating from both sides. By embedding 6 prongs in Variant X4, adhesion and heating eﬃciency are increased even further. The focus in Variant X5 was on decreasing the spring constant by lengthening the silicon joint of Variant X3.

Both X4 and X5 are in eﬀect silicon-BCB bimorph actuators. The stiﬀness of silicon

BCB heated

Si wafer with etched groove BCB cold (deflected due to shrinkage during curing)

a g/2

a b w

( )

g=2 atan (b-a) 2w

Figure 1: Principle of a V-shaped polymer actuator.

(22)

Out-of-plane: In-plane:

Si wafer with

KOH etched groove DRIE etched in- plane structure

BCBBCB BCB

Figure 2: Principle for in- and out-of-plane actuation.

is much higher than that of BCB, so the expansion or contraction of the BCB is transformed to a rotation.

The non-linearity of the actuator, i.e. the oﬀ-axis deﬂection caused by the circular movement (illustrated in Figure 5), can be handled by a control mechanism. This control, however, quickly becomes more complex as the degrees of freedom increase, making a solution that compensates for this non-linearity a viable alternative. A proposed design, shown in Figure 4(a-b), combines 4 actuators in such a way that they cancel out each other’s non-linearity. Figure4(c-d) illustrates how two of these 4-actuator constellations can work together to create a two-dimensional in-plane positioning platform with uncoupled movement.

2.5 Dimensions

The goal of the fabricated devices was to prove the concept with a single actuator.

Although it is possible to place several actuators in series to increase the angular span of the actuator and decrease the required lever length, the linearity of the movement suffers, and the crosstalk between the different degrees of freedom is increased. There- fore, working with the assumed optical component size of 1 mm³, a lever length of the same order of magnitude (1.5 mm) was selected. Figure5illustrates the crosstalk at three different deflections for lever lengths of 0.5 to 2 mm, an ideal hinge actua-

b) X2

c) X3 e) X5

a) X1

d) X4

390 µm 620 µm

310 µm 290 µm

Polysilicon heating resistors on Si prongs

200 µm

420 µm

Si BCB

Polysilicon heating resistors

Figure 3: Top view of the evaluated in-plane actuators.

(23)

(a) (b)

(c) (d)

Figure 4: (a-b) Combination of 4 actuators with springs to achieve parallel actuation: (a) before release; (b) after release; (c-d) two-dimensional in-plane positioning platform with near uncoupled movement.

tor being assumed. At 40 µm total deflection, the crosstalk increases significantly for levers shorter than 1 mm. For lower maximum deflections, the non-linearity decreases rapidly, and shorter levers can be designed.

The width of the in-plane actuators was chosen to be wider than the out-of-plane actuators to compensate in part for the reduced volume, but it was not extended to more than 200 µm in a conservative move of not straying too far from the proven concept. Wider actuators would have strengthened them both in the sense of the force they could deliver and the weight they could carry.

The main factor in selecting the height of the actuators was the possibility of covering the actuators with BCB. Since the BCB needs to cover partially the edge of the actuator, problems could occur if this edge becomes too thick. Simulations were used to dimension the 4-actuator constellation and predict its displacement.

ANSYS [43] was used to design Variant X1.

2.6 Performance

First, all of the variants were evaluated with external heating. In an automated measurement setup for the in-plane actuators, the actuator was placed on a hotplate and the deﬂection was photographed and logged as a function of the temperature as it was being ramped up and down. The image data were then analyzed with Matlab [44]

(24)

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 0

1 2 3 4 5 6 7 8 9 10 11

Lever length, L (mm) Relative down movement, dy/dx (%)

d_x = 40 m

d_x = 20 m d_x = 10 m d_x = 40 m

d_x = 20 m d_x = 10 m d_x = 40

d_x = 20 m m

d_x = 10 m

BCB

L

Lever arm

Heater dy

d_x

m m

1

2

÷÷ø- ççè ö - æ

= x x

x

y dL

dL d d

Figure 5: Plot indicating the relative non-linearity (dy/dx) as a function of the lever length (L) and maximum deﬂection (dx). The inset formula provides a mathematical description.

to generate the measurement curves presented in Figure6a. The deflection is indicated both as the measured deflection (on the right Y-axis) and in normalized units (left Y-axis). Measurements on the Y-actuator (out of plane) were performed by aiming a laser at the actuated surface and measuring the movement of the reflected beam using a camera.

Three of the in-plane actuators were evaluated using the integrated heaters (Fig- ure 6b). These measurements were performed in a similar way as those shown in Figure6a, except that the voltage over the integrated heater was used to actuate it instead of the hotplate. The non-linearity of the curves for X2 and X3 is due to measurement errors during early evaluation, i.e. before the development of the automated measurements, but after destructive testing of the actuator types involved.

The deflection caused by the weight of the platform (1.2 mg) was measured by placing the test platforms at different orientations. Due to the uncertainty in the measurements, which was around 5 µm, no deflection was noted in any direction.

This provides a lower limit of 4 GPa for Young’s modulus, which conﬁrms theoretical values. FEMLAB (Comsol AB, Stockholm, Sweden) software was used to estimate the spring constant of the devices.

2.7 Discussion and outlook

Now that the feasibility of using BCB as an optically compatible actuator material for both in-plane and out-of-plane actuation using integrated heating has been demonstrated, a number of design improvements may be suggested.

In the original design, two contact lead feed-through methods were evaluated:

(25)

Temperature (°C) Deflection (mrad) (µm deflection / mm lever arm)

40 60 80 100 120 140 160

-2 0 2 4 6 8 10 12 14

16 X5, 0.11 mrad/K X4, 0.09 mrad/K X1, 0.06 mrad/K X3, 0.03 mrad/K X2, 0.03 mrad/K Y, 0.03 mrad/K

40 60 80 100 120 140 160-5

0 5 10 15 20 25 30 35 40

Actual deflection on test platform (µm)

Power (W)

0 0.1 0.2 0.3 0.4

0 1 2 3 4

5 X3, 16 mrad/W X2, 13 mrad/W X1, 3.5 mrad/W

0 0.1 0.2 0.3 0.40

2 4 6 8 10 12 14

Actual deflection on test platform (µm)

Deflection (mrad) (µm deflection / mm lever arm)

(a) (b)

Figure 6: (a) Temperature - deﬂection curve, (b) power - deﬂection curve.

passing aluminum leads over the actuators and passing polysilicon leads through spring structures alongside the actuators. The ﬁrst alternative, besides impeding the actuator’s movement, added unnecessary diﬃculties to the fabrication and was abandoned at an early stage. Measurements on X3 indicated that adding spring-like structures does not overly impede the movement and the relative counteracting force would be reduced with wider actuators. Having the leads go past the actuator is necessary to achieve more than single-axis actuation and also has the advantage of providing heating from both sides.

Another conclusion that can be drawn from the diﬀerence between single-sided and double-sided heating (X1 vs X3) is that BCB also acts as a good thermal insulator.

The addition of passive elements would reduce thermal crosstalk without signiﬁcantly reducing structural integrity. Such passive elements would be rectangular BCB areas, which would not generate any torsional movement. Combining active and passive elements, one can reduce the thermal mass needed to heat up the actuator and control the heat conduction between independent actuators.

For applications where the weight of the moved object is less, fully uncoupled movement can be achieved using the design in Figure4. The reason for this weight limitation is that the spring structures have to be weak enough to allow bending by the 4 actuators. The results of the preliminary tests of the construction using external heating are promising.

(26)

(27)

3 MEMS MICROMIRROR ARRAYS 13

3 MEMS micromirror arrays

3.1 Introduction

Micromachined mirror structures have been fabricated for over 25 years, the technology being driven by the high quality of these structures and the promise of low fabrication costs. That micromachining easily lends itself to parallel production makes micromirror arrays feasible. In array format, micromirrors act as light modulators that can modulate the phase or amplitude of incident light [45,46,47,48,49]. The main application areas of micromirror arrays today are projection display systems [50], pattern generators in maskless lithography systems [51,52,53,54], optical scanners [55], laser printers, optical spectroscopy [56], aberration correction [57], adaptive optical systems [58,59], and switches and cross connectors in optical communication systems [57]. Large two-dimensional arrays of individually addressable micromirrors require on-chip electronics. The intended application of the micromirror arrays in this chapter, microlithography, is an application that requires individually addressable micromirrors. The micromirrors should be analog controllable for optimum resolution, unlike, for example, the digitally controlled micromirror arrays used for projection displays developed by Texas Instruments Inc. [50].

3.2 Maskless microlithography

Microlithography is the process of transferring a pattern from a photomask onto, for example, a silicon wafer coated with photoresist. In the microelectronics industry, steppers are used to project the pattern several times from the reduced photomask onto the wafer. In the electronic packaging and the ﬂat-panel display industry, similar machines, called aligners, project the full pattern at once onto the substrate.

The state-of-the-art photomask sets used by these systems are produced by directed electron-beam writing for the highest resolutions and by scanning laser for the less critical resolutions. Both are slow and expensive methods that can drive reticle costs for a mask set up to USD 500,000 [60]. This cost, which is expected to increase dra- matically as critical dimensions decrease, is a major driving force in the search for alternative ways to produce photomasks.

In 2000, Micronic, a leading manufacturer of scanning-laser based pattern gener- ation equipment, introduced a new system based on an array of micromirrors that acts as a spatial light modulator (SLM). The principle of this system, which Micronic developed together with the Fraunhofer Institute IPMS, is illustrated in Figure7a.

Coherent light is created by a laser and impinges through a beam splitter and a lens system onto the micromirror array. Depending on the deflection of the micromirrors, one can control the amount of interference at the focal plane of the Fourier lens.¹ By filtering the Fourier transformation of the light using an aperture, only the 0^th mode is passed through. When another lens performs the inverse Fourier transformation, the pixel’s gray scale can be controlled from fully illuminated (some flat mirrors) to

1A Fourier lens is a normal convex lens used in a distinct way: when it refracts a parallel bundle of rays, the intensity of the pattern formed at the focal plane of the lens is related to the square of the amplitude of the Fourier transformation of the input signal. If the focal plane coincides with the focal plane of another convex lens, the lens will produce an inverse Fourier transformation.

(28)

0

-1 +1

F F'

laser light deflected

mirrors SLM

non-deflected mirrors

Fourier lens, LF

projection optics, LP

beam splitter

image plane bright

spot dark spot Fourier plane

spatial filter fF

fP

(a)

Pulsed excimer laser Interferometer

SLM Pattern rasterizer

Aperture Lens

Mask Y-position

Beam-splitter

Lens

Laser Interferometer

X-position

(b)

Figure 7: (a) Principle of the spatial light modulator (from [61], with kind permission of U.

Dauderst¨adt and P. D¨urr, Fraunhofer IPMS); (b) the SLM in a mask writer setup (courtesy of Micronic AB, Sweden).

fully obscured (some mirrors at deﬂection d = λ/4). Analog control of the deﬂection of the mirrors enables the gray scale to be controlled very precisely to achieve very high resolution.

To address the mirrors individually, the array must be integrated onto CMOS circuitry. The solution that Micronic and the Fraunhofer Institute proposed is to use a low-temperature process to integrate aluminum mirrors onto the circuitry [61].

With it, they were able to produce high-resolution photomasks at a higher rate, which created the possibility of also using the SLM for maskless lithography applications.

Maskless lithography is used, for example, when one wishes to fabricate application- speciﬁc integrated circuits (ASICs) or prototype CMOS wafers without paying for an entire high-end mask set.

3.3 Monocrystalline silicon as a mirror material

The most common structural materials for micromirrors are aluminum, polycrystalline silicon, and monocrystalline silicon. Monocrystalline materials have several advantages because of their superior mechanical stability relative to metals. In contrast to metal micromirror hinges, there are no recrystallization effects due to material deformation in monocrystalline silicon hinges. This minimizes hysteresis and memory effects, which is very important in applications where the mirrors must be actuated to a number of discrete tilting angles (analog addressing) [52,54]. In addition, the achievable optical quality – the surface roughness and the uniformity of monocrystalline silicon surfaces – is superior to most other surfaces [45,55,59]. It is extremely difficult to fabricate completely flat metal mirrors due to the stress gradients that occur when thin metal films are deposited by electron-beam evaporation or sputter- ing [62].

(29)

The integration of micromirrors and integrated circuits is most commonly achieved by means of monolithic integration of the micromirrors [47,50,51,52,53,54]. The thermal budget of CMOS, however, prevents deposition of high-quality materials such as monocrystalline silicon since they require high-temperature processes. This is why assembly of micromirrors on CMOS wafers is an attractive technology for manufacturing monocrystalline silicon micromirror arrays.

3.4 Wafer-level assembly and its application in the fabrication of micromirror arrays

To circumvent this problem, hybrid integration techniques such as ﬂip-chip bonding or compression bonding [63,64] have been used. However, these techniques have severe limitations in terms of minimum achievable feature size and thickness of the micromirrors and in the alignment and distance control between the micromirrors and the electrodes on the integrated circuits. In this project, we adapted a technique called membrane transfer bonding [65], which has been used to create polycrystalline silicon bolometers. It is a tool for the transfer of very thin membranes at low temperatures.

In membrane transfer bonding, the material one wants to put on the target wafer is transferred from a carrier wafer by polymer adhesive wafer bonding.

This process, shown in Figure 8, is started by spinning a photoresist layer on a prepared target wafer. This photoresist is specially selected for its property of not outgassing above a certain temperature (in this case 70^◦C), while retaining its

“stickiness”. This is important when it is used as an adhesive bonding layer, since outgassing would generate bubbles, which are detrimental to the bond quality. After this, the sacriﬁcial wafer with the monocrystalline layer on top of an oxidized carrier wafer is bonded to the target wafer. The carrier wafer and buried oxide are then removed by a combination of grinding and reactive ion etching (RIE): deep RIE for the silicon and ordinary RIE for the silicon dioxide. No precision bond alignment is needed, since the monocrystalline layer is patterned and etched only after the removal

Polymer bond with Ultra-i 310 resist

Remove sacrificial wafer

Pattern &

etch mirrors

Remove resist in O₂ plasma Fix mirrors

using electro- plating Sacrificial wafer

(SOI-wafer)

Target wafer (e.g. CMOS-wafer) Addressing electrodes (Al)

Target Wafer (e.g. CMOS-Wafer)

Sacrificial Wafer (SOI-Wafer) Etch-stop layer (SiO₂)

Mirror membrane (monocrystalline Si)

Post (Au) Adhesive material

Contact layer (Au)

Target Wafer (e.g. CMOS-Wafer) Distance holder (Si₃N₄) Insulation layer (Si₃N₄)

Figure 8: The membrane transfer bonding technique used for the fabrication of micromirror arrays.

(30)

Mirror thickness: 0.34 µm Gap height: 0.8 µm Distance holders: 0.2 µm Monocrystalline

silicon mirror

Distance holders (Si₃N₄)

Posts (Au) Electrodes (Al)

Figure 9: A torsional micromirror.

of the wafer. The mirror membranes are ﬁxed to the substrate by electroplating gold up from gold posts to form rivets. Finally, the bonding resist is removed with an oxygen plasma.

During the entire procedure, the temperature never exceeds 110^◦C, so the CMOS circuitry is not damaged. Since the monocrystalline layer is transferred before it is patterned, the alignment precision and patterning resolution of the mirrors are only restricted by the limitations of photolithography. This is a signiﬁcant advantage, since bonding alignment with current technology is limited to 2-3 µm precision, which would be unacceptable for this application.

3.5 Design

Figure 9 shows a schematic drawing of the torsional micromirror. It consists of a monocrystalline silicon membrane that is connected to a silicon substrate by two gold posts. Aluminum addressing electrodes are formed on the silicon substrate below the mirror plate. The mirror plate is electrically connected via the gold posts to the gold metallization layer on the silicon wafer. To electrostatically actuate the mirror, a voltage is applied between the mirror membrane and one of the two addressing electrodes. Consequently, the electric ﬁeld actuates the silicon mirror like a seesaw, and the mirror hinges act as linear torsional springs. As a result, typical deformation characteristics of torsional micromirrors are strongly non-linear, and a voltage exists at which the mirror is suddenly pulled in. Distance holders prevent the mirror membrane from contacting the addressing electrodes when the pull-in voltage is exceeded. A pulled-in mirror membrane is released again at a much smaller voltage than the pull- in voltage. A detailed study of the design parameters and the characteristics of torsional micromirrors is given elsewhere [66,67]. To demonstrate the concept, we designed arrays of 4× 4 monocrystalline silicon micromirrors that can be addressed and actuated line by line. The mirror membranes are 0.34 µm thick and measure 16 µm× 16 µm.

3.6 Discussion and outlook

This chapter presents the proof of concept for integrating high-quality monocrystalline mirror membranes onto CMOS circuitry. The development of the membrane transfer

(31)

method has offered not only the fabrication of micromirrors but also in general the integration of high-quality materials without the limitations imposed by the thermal budget of CMOS. That the process is applicable on wafer scale makes the principle attractive for production. Specifically for the application of micromirror arrays, the next step is to go from the research stage to the development stage. The mirrors were actuated row by row, using a simple contact lead structure. Integrated circuitry will have to be developed, and larger arrays will have to be fabricated. The most critical aspect here will be the fine tuning of the bonding process to approximate 100% yield.

(32)

(33)

4 ON-CHIP DNA ANALYSIS 19

4 On-chip DNA analysis

4.1 DNA and single nucleotide polymorphisms

Fifty years after the discovery of the structure of DNA, DNA analysis has grown to become one of the largest fields in science, and the completion of the mapping of the human genome has moved the focus from exploring to understanding. The invention of the polymerase chain reaction (PCR) in 1983 enabled biochemists to multiply DNA theoretically indefinitely. With applications in disease research, drug development, medical diagnostics, forensics and agriculture, huge amounts of data have to be processed. The basic tools for DNA analysis exist, but new tools and technologies need to be developed to make the analysis faster and cheaper. One specific aspect of DNA analysis is the identification of single nucleotide polymorphisms (SNPs). A SNP (pronounced snip) is a small change or variation in a single base in a DNA segment. A variation must occur in at least 1% of the population for it to be considered a SNP. SNPs make up 90% of all human genetic variations, and occur every 100 to 1000 bases along the human genome [68,69]. The importance of SNP scoring in medical and other applications is that a SNP occurring in a segment of coding DNA (the 3-5% of the DNA that codes proteins) can alter the biological function of a protein and affect a person’s vulnerability to a disease or reaction to medication.

4.2 SNP scoring

In many applications, the SNP position of interest is already known, and the goal is to determine the frequency of a speciﬁc SNP occurrence in a given population. This process is called SNP scoring. The majority of SNP analysis chemistries in use today are based on a method called allele-speciﬁc oligonucleotide hybridization²(ASOH) [70,71].

This protocol relies on using hybridization to distinguish between two DNA molecules differing by one base. Oligonucleotides (short sequences of around 20 nucleotides) rep- resenting SNP sequences are hybridized (bound) to the PCR fragments. Fluorescent labels that fluoresce more intensely when the compound is hybridized are attached to the compound. After hybridization and thorough washing, fluorescence intensity is measured for each SNP oligonucleotide. If the oligonucleotide matches the PCR fragment, the fluorescence intensity will be high due to the high concentration of hybridized compounds. In the case of a mismatch, the pair will have denaturated,³ or hybridization will not have occurred, and there will be very little fluorescence.

4.3 Dynamic allele-specific hybridization

Allele-specific oligonucleotide hybridization (ASOH) in its basic form is limited by the difficult challenge of defining discriminatory assay conditions for a series of simul- taneous analyses on a given microarray. Therefore, newer methods with additional steps such as enzymatic reactions on the oligonucleotide arrays are used to improve the discriminatory power of microarray-based SNP analysis [72,73,74,75]. However, these methods also entail considerable optimization efforts and/or costly enzymatic

2Hybridization: the process of joining two single strands of DNA.

3Denaturation: the process of splitting double-stranded DNA.

(34)

ROX: attached to the 3' end of the probe. Acts here as the acceptor in the iFRET

Single strand of DNA (55 bases) Oligonucleotide probe (17 bases)

Biotin Streptavidin

or

Sybr green: an intercalating dye, i.e. a dye which attaches to double-stranded DNA. Acts here as the donor in the iFRET.

T C G G C

C G

A

Match 3' end

5' end

Mismatch

T C G G G

C G

A

Figure 10: Schematic representation of a DNA-probe duplex for DASH analysis.

or oligonucleotide labeling steps. One recently developed method for SNP scoring that has shown promise is dynamic allele-speciﬁc hybridization (DASH). The essence of DASH is the same as that of ASOH, but it enhances the discriminatory power by monitoring the DNA denaturation under dynamic heating and does not suﬀer from the same drawbacks: no additional enzymatic reactions or optimization steps are needed.

Figure 10shows a PCR/probe duplex, as it appears in the DASH protocol. The DASH principle is shown in Figure11. The DNA/probe duplex is heated to a temperature above its denaturation temperature, which is also called its melting temperature. Upon denaturation, the ﬂuorescence of the duplex decreases. For a matching DNA/probe duplex, the melting temperature will lie higher than for a mismatching one, causing the ﬂuorescence to decrease around that temperature (see Figure11a).

To discern the match from the mismatch easily, the negative derivative of this curve is plotted as shown in Figure11b. Since humans have two chromosomes, there is always the possibility that both a match and a mismatch occur, the so-called heterozygous case as opposed to the homozygous case in which both alleles are the same. This case is readily identiﬁed by the double bump in the negative derivative curve. The following paragraph will give a short background to the ﬂuorescence method used in this work.

4.4 Fluorescent detection methods

Fluorescence is used to measure the denaturation of the DNA/probe duplex in DASH.

This use is widespread in biochemistry: there are a large number of ﬂuorescent dyes

(35)

(a) Temperature

Fluorescence

He tero

zyg ous Match Mism

atch

(b)

T

C G

G G CG

A

Mismatch

T

C G

G C CG

A

Match

Temperature

-dF/dT

Figure 11: The DASH procedure, (a) shows the measured ﬂuorescence decrease, (b) shows the negative derivatives of the curves in (a).

that, when illuminated at one wavelength, emit light at another, longer, wavelength.

By attaching dyes to molecules, they can easily be followed by fluorescence micro- scopes that are equipped with filters to select specific wavelengths. Many methods for SNP analysis are based on whether or not the studied PCR fragment is hybridized with a probe.

The most straightforward way to detect the denaturation of DNA is to have the PCR-generated strand attached to a solid phase (e.g. a surface) and to tag the probe with a fluorophore⁴(see Figure 12a). When the DNA duplex is hybridized, a strong fluorescent signal can be detected from the solid phase. When the duplex denaturates, the probe with the fluorophore diffuses away from the solid phase into the mixture, and the fluorescent signal loses intensity. Another common way of sensing the state of DNA is to use an intercalating dye, as illustrated in Figure12b. This is a dye that specifically binds to double-stranded DNA (dsDNA) causing a surface coated with it to fluoresce much more than one coated with single stranded DNA (ssDNA). This second method is cheaper since it does not require the binding of a fluorophore to the probe, but it also suffers from the presence of a significant background signal, even if the DNA is not hybridized.

A third detection method is called fluorescence resonant energy transfer (FRET, Figure 12c) in which one strand has a donor molecule attached and the other an acceptor molecule. The acceptor fluorophore is excited by the light emitted by the donor fluorophore, so it only emits light when it is close enough to the donor. The efficiency of the energy transfer between donor and acceptor is inversely proportional to the sixth power of the distance between them, which makes this an extremely sensitive method. The FRET can then be detected by measuring the appearance of acceptor fluorescence or the quenching of donor fluorescence. The drawbacks of this

4A fluorophore is a fluorescent molecule, i.e. a molecule that, upon absorbing light of a certain wavelength, emits light of a longer wavelength.

(36)

A

D D

A

A b) Intercalating dye:

c) FRET:

d) iFRET:

Denature

F F

a) Fluorescent tag:

Denature

Figure 12: Four methods for detecting DNA denaturation (partly from [76], with kind permission of Mathias Howell).

method with respect to the use of an intercalating dye are the cost of attaching the two ﬂuorophores and the much weaker ﬂuorescence intensity.

A method that combines the advantages of the methods based on intercalating dye and FRET is called induced FRET (iFRET [76]), as shown in Figure12c. The FRET is induced by the presence of the intercalating dye. This provides the contrast and low background signal from FRET with the lower cost intercalating dye. This ﬂuorescence method is used for conventional DASH as well as for the miniaturized version described here.

4.5 Miniaturization

The main focus in the development of DASH from an academic concept to a com- mercial success is, as for all the other SNP scoring systems, to be able to perform a large number of scoring experiments with as little reagent as possible in as short a time as possible. Using instruments from conventional molecular biology, the DASH assay has evolved from 96-well microtiter plate⁵ single-plex sample processing, manual pipetting and intercalating dye ﬂuorescence signals to 1536-well multiplex sample processing with automated pipetting and iFRET detection. For handling even smaller amounts, microsystem technology is the next logical step. Although miniaturizing a biological assay is not straightforward, DASH lends itself well to miniaturization. The need for precise and rapid temperature control of the reagent is a factor that is more easily controlled in microscale than in macroscale. The idea of using microfabricated

5Microtiter plates are plastic sample holders with typically 96, 384 or 1536 sample wells arranged in a 2:3 rectangular matrix.

(37)

systems to perform on-chip genomic analysis is a step towards the creation of micro total analysis systems (µTAS).

4.6 Microspheres

A helpful tool in the miniaturization of surface chemistry has been the use of microspheres (beads). These beads, which have diameters ranging from 20 nm up to 1 mm, are traditionally used in suspensions, where they act as a solid phase on which the reagents can react. They can be made of a variety of materials, ranging from all kinds of polymers to silica, and can contain different levels of magnetite to allow manipulation by subjecting them to magnetic fields. Their surface can be coated with proteins or antibodies or functionalized to bind to reagents or surfaces. They exist in different colors for easy identification even by automated systems and can be dyed with fluorescents.

For microsystems, beads can be used to increase the reaction surface and to control where the reaction takes place. They can be physically trapped in microﬂuidic systems by cage-like structures [77], held in place or moved by local magnetic ﬁelds [78,79,80]

or fixated using surface chemistry [81,82]. Parallelization of reactions, also called multiplexing, can be done by precoating different bead types with different reagents, mixing them, and performing the same reaction steps on them. The results can be distinguished by different ways of fluorescently tagging the beads or by using different bead sizes.

A popular technique for immobilizing beads on a surface is the streptavidin/biotin system. This system has one of the strongest bindings known for the non-covalent binding of a protein (streptavidin) and a small ligand (biotin). Streptavidin-coated beads are commercially available. It is relatively easy to coat surfaces with biotin and bind streptavidin-coated beads to them. DNA oligonucleotides or PCR products can also be biotinylated to permit the binding of DNA strands to the beads.

4.7 DASH on chip

This section will describe the manufacturing of the chip, the sample preparation and the measurements. The description is kept on a conceptual level, as the details (exact layer thicknesses, concentrations etc.) can be found in Paper 7.

4.7.1 Bead monolayers

In a preparatory study [83], it was shown that it is possible to create a pattern of biotin on a surface by printing with PDMS stamps and to attach streptavidin-coated beads selectively to it. In this way, monolayers of beads can be created. The bond was strong enough to withstand the alkali treatment, the heating and the convective forces of the liquid acting on the beads. With this knowledge, we designed heater chips that could be patterned with beads.

Assembly of microsystems for optical and fluidic applications