Assembly of microsystems - CiteSeerX

Keynote Paper

Annals of the CIRP, Vol. 49 (2), 2000, pp. 451-472

Assembly of microsystems H. Van Brussel (1), J. Peirs, D. Reynaerts, A. Delchambre, G. Reinhart (2), N. Roth (2), M. Weck (1), E. Zussman (2)

Abstract In the microworld, as well as in the macroworld, assembly is a crucial operation in the genesis of a product. This keynote paper focusses on the assembly problems occurring in the manufacturing cycle of microsystems. Scaling effects make that the assembly problems are different in the microworld. The different assembly operations and techniques, like manipulation by physical contact, non-contact manipulation, smart assembly techniques, and joining methods are thoroughly discussed. Finally, some relevant examples of micro-assembly systems and of assembled microproducts are given. Keywords: Assembly systems, Miniaturisation, Micro-electromechanical systems

1 TERMS OF REFERENCE Where natural creatures grow from a single cell, manmade artifacts, like industrial products, are mostly assembled from different components. Assembly is therefore a very important process in the genesis of a product. There are many reasons why products may consist of different parts. Different functions require different materials in one and the same device. In mechanisms, pivots require flexible materials while the links themselves must be stiff. Electrical actuators are made of conducting as well as isolating materials. The rotor and the stator of a motor are different parts by their very nature. Products may be too complex to be produced as a single part. Some parts are wearing out and must be regularly replaced. The abovementioned arguments are true for macro- as well as for microproducts. This keynote paper focusses on assembly problems occurring in the manufacturing cycle of microsystems. For our considerations, microproducts have dimensions in the millimetre range and below. We do not include structures in the nanometre range. Miniaturisation of electromechanical systems is a hot issue in modern technology. The resulting products, called micro-electromechanical systems (MEMS), are claimed to have a vast – yet virtually untapped – potential. Although the field is still to mature, a multibillion dollar market is predicted [1]. It therefore belongs to the mission of a scientific organisation like CIRP to closely monitor the evolution of this emerging technology. A quick survey of past publications, in the CIRP Annals, on manufacturing issues of MEMS yields only a small harvest [65,66,76,95,97,98,100]. The many contributions, from CIRP members, to this keynote paper however indicate a vivid interest in microsystems technology (MST) within the CIRP community. To miniaturise a product it is not sufficient to simply reduce its dimensions. A lot of new problems emerge, related to scaling effects, manufacturing problems and, of course, assembly problems. One of those problems has

to do with the fact that, up to now, nearly all production techniques for microsystems have their origin in the microelectronics technology, and are essentially 2Dprocesses. Assembly is required to realise 3D-parts out of parts made by those 2D manufacturing methods. Assembly further allows the manufacture of objects consisting of parts requiring incompatible technologies. For instance, processing of GaAs is not compatible with Si processing, especially with the high-temperature oxidation and dopant diffusion steps that destroy the GaAs material and contaminate the Si circuitry and furnaces. Through an assembly step, GaAs optoelectronic devices can be integrated on Si VLSI circuitry. Assembly also allows the choice of the most optimal production method for each part. For instance, it is difficult to produce high-quality-factor coils on-chip, mainly due to the difficulty to create the third dimension on-chip. By producing these coils by conventional winding and by assembling coil and actuator afterwards, both quality and production cost can be improved. The paper is organised as follows. The typical features of assembly in the microworld are outlined in section 3. Section 4 classifies the assembly systems relevant for microsystems. Manipulation issues are crucial to assembly ; manipulation with physical contact is explained in section 5, while section 6 deals with noncontact manipulation. Some so-called ‘smart’ assembly techniques, particular for microsystems, are dealt with in section 7. Joining of microparts is considered in section 8. Some industrial examples of micro-assembly systems and of assembled microproducts are given in sections 9 and 10 respectively. Some closing remarks conclude the paper. 2 MARKET PROSPECTS FOR MEMS AND MST [2] In 1998, NEXUS (The European Network of Excellence in Multifunctional Microsystems) established a task force entitled ‘Market Analysis MST’ to prepare an applicationsoriented in-depth analysis of MST markets from 1996 through the year 2002. The total world market for 451

microsystems is expected to grow from 14.4 billion Euro in 1996, to 38 billion Euro by the year 2002. This reflects growth rates of 18% per year. The top seven markets in 2002 are predicted to be: • Hard disk drive heads (12 billion Euro) • Inkjet printer heads (10 billion Euro) • Heart pace makers (3.7 billion Euro) • In-vitro diagnostic devices (2.8 billion Euro) • Hearing aids (2 billion Euro) • Pressure sensors (1.3 billion Euro) • Chemical sensors (0.8 billion Euro) Six main application domains have been identified: • IT peripherals • Medical/biomedical applications • Industry and automation (including aerospace) • Automotive applications • Environmental monitoring Products that have a high probability of being on the market by 2002 are found to be: • Drug-delivery systems • Optical switches • Lab-on-chip systems (DNA, HPLC, ..) • Magneto-optical heads • Projection light valves • Coil-on-chip • Micro-relays • Micromotors Integrating silicon micromachining with CMOS technology remains difficult, while hybrid (multichip) microsystems have gained in importance. Microstructuring, not only in silicon, but also in polymers, metals, and ceramics has been enabled by advances in microfabrication techniques. These latter include laser machining, highaspect ratio microreplication based on lithographic patterning, electrodischarge machining (EDM), diamond milling and other precision mechanical removal processes. There is a need for fully-3D batch production micromachining processes. Micro-EDM, combined with micro die casting may provide a solution. 3 ASSEMBLY PROBLEMS IN THE MICROWORLD The main difference between macro- and micro-assembly is the required positional accuracy of automatic assembly machines. In the macroworld, a precision of a few hundred microns is typical for serial link robotic manipulators with four to six axes. In the microworld, submicron precision is often required, comparable to wafer stepper precision. This degree of precision is beyond the calibration range of conventional open-loop precision assembly devices used in industry. Closed-loop strategies are required to compensate for poor kinematic models and thermal effects. Real-time vision feedback is perfectly suited for this application. Moreover, a manipulator should have a 3D workspace instead of the wafer stepper’s 2D workspace, making the manipulation problem considerably harder. Obtaining accurate sensor information to close the loop is also difficult as sensors can be too bulky to be placed on tiny precision instruments and they have to be extremely sensitive as forces and displacements are very small. The alternative of image processing has its problems too: it is slow, costly, difficult to program, and susceptible to reflection, light condition, colour changes, ... Moreover, the view may be obstructed by tools that are orders of magnitude larger than the parts being handled, or the lenses, cameras and other optical instruments may obstruct proper manipulation of the object, especially as multiple optical axes are required for real 3D manipulation. Thus

to enable micromanipulation, miniature sensitive sensors and extremely accurate 3D robotic manipulators have to be built. A second major difference between assembly in microand macrodomains is the mechanics of object interactions. In the macroworld, the mechanics of manipulation are predictable, e.g. when a gripper opens, gravity causes the part to drop. In the microworld, forces other than gravity dominate due to scaling effects. Surface-related forces, such as electrostatic, van der Waals and surface tension forces become dominant over 3 gravitational forces. Mass decreases with L while stiffness for bending and tensile strength are proportional 2 to L and L respectively. Due to this unevenly scaling behaviour, manipulation in the microworld is completely different from manipulation in the macroworld. Manipulation in this ‘strange’ world, therefore, requires training of the human operator. A difficult problem in manual handling in the microworld is the loss of direct hand-eye co-ordination. The microscopes and tools limit the ability to directly see and sense the objects to be handled. The tools used to manipulate the objects have less degrees of freedom than the human hand and there is no force feedback. The operator’s view is restricted for a number of reasons. First, the high magnification restricts the view to a very small area, such that the operator lacks global information about the object. Therefore, variable magnification in a wide range is indispensable. Second, in an optical microscope with very high magnification, the depth of focus is often very short in comparison with the width of the field of view. The limited depth of view impedes clear images of non-planar objects or moving or vibrating structures because the structure moves in and out the focussed plane. Third, the working distance (the distance between objective lens and object) also tends to be short. This hinders manipulation of objects and tools. Weck et al. [3] built an assembly system in the vacuum chamber of a large-chamber scanning electron 3 microscope, with a volume of nearly 2 m . The main disadvantages of this solution are its high price and the long evacuation time. A big issue is also the trade off between field of view and resolution. An elegant solution of this problem is the so called ‘eye on the hand’, i.e. a camera is mounted on the arm of the robot [100] or in the gripper [3]. The main advantage of this solution is that the tool centre point is always in focus. The main challenge is then to fabricate a small-sized camera with a lean structure. Weck et al. [3] solved this problem by integrating a miniature endoscope into the gripper. Finally also the cost of the manipulation has to be considered. Most microparts are produced in batch processes, with hundreds or thousands of them on a single wafer. This massively parallel production mode is the main factor for cost reduction. This cost reduction is one of the major reasons for microsystems to replace their macroscale counterparts. When assembly of these microsystems is performed one by one, be it manually or automated, it will considerably increase the production cost. Therefore, massively parallel micro-assembly systems may be required. 4 MICRO-ASSEMBLY SYSTEMS The dimensional and physical mismatch between the micro- and macroworld and the required precision, operator stress and eye strain associated with assembling minute parts under a microscope ask for advanced micro-assembly tools. 452

4.1 Master-slave systems As the assembly and grasping forces are too small to be sensed by a human operator, manual assembly with tweezers is performed purely based on visual feedback. In the macroscopic world a human operator uses vision for coarse and non-contact servoing and combines it with force feedback for accurate positioning under contact. A good micro-assembly system should measure these extremely small forces with microsensors and feed it back to the human operator. A master-slave macro-micro teleoperation system has been built and tested by Kaneko et al. [4]. The position of the master arm, manipulated by the operator, is scaled down and used to control the position of the slave. The forces measured by the slave are amplified and applied to the master arm through some transfer function. How to set up these transfer function gains to provide the operator with a natural sensation remains to be investigated. From an intuitive point of view however, force scaling should not be dynamic as this introduces lead or lag between input and output signals. This would make it difficult for the operator to telemanipulate the objects with real-time sensation, and could cause the slave to break or damage the objects. Figure 1 illustrates the layout of a tele-micro-surgical system used in an experiment of suturing an artificial blood vessel of 1 mm diameter [ 5].

Figure 1: Tele-micro-surgical system [5]. 4.2 Automatic assembly machines To reduce the assembly cost, automatic micro-assembly machines might be a good solution. Zhou et al. [6] use a combination of visual and force feedback to control the grasping force. Visual servoing allows for controlled motion at mm/s speeds and submicron repeatability. However, nanometre repeatability cannot be achieved with visual feedback alone. Force sensing yields much higher resolution, but works only when in contact with the object. A combination of both force and visual feedback allows fast movement in free space without the danger of large impact forces when the contact is made. Contact forces of 2 nN with impact forces of 9 nN were achieved for micropart approach velocities of 80 µm/s. Feddema et al. [7] use a CAD-driven technique, where the objects, their position and orientation are recognised by comparison with a synthetic image from CAD. Diffraction and out-of-focus effects are added to the synthetic image to make it more resemble the expected real image. A modular micro-assembly system with 4 degrees of freedom is shown in figure 2 [8]. It consists of an x-y positioning table, above which an overhead manipulator and a stereomicroscope with a CCD camera are located. The overhead manipulator can move along the z-axis and rotate around this axis; it can be equipped with different grippers or applicators by an automated turret tool changer. The positioning accuracy of the micro-assembly

system is 2 µm. It is used to assemble micro-optical duplexers, which consist of two 0.9 mm spherical lenses, 3 a 3x3x1 mm wavelength filter, and a glass fibre cable.

Figure 2: Modular micro-assembly system [8]. 4.3 Assembly by microrobots As tolerances in micromechanics lie in the nanometre range, assembly robots should be very precise. The manipulation accuracy of conventional robots is mechanically limited, since disturbing influences which are often negligible in the macroworld, such as fabrication defects, friction, thermal expansion or computational errors, play an important role in the microworld. Furthermore, these robots are subject to mechanical wear, and must undergo regular maintenance and calibration, which makes them expensive. As an alternative, Fatikow [9][10] proposes microassembly with microrobots. Figure 3 illustrates the approach. The microrobots, 50 to 80 mm in size, stand on piezoelectric legs and move based on the stick-slip principle. This allows very fine resolution down to 10 nm, while speeds up to 30 mm/s can be obtained. The robots move on a glass plate with three degrees of freedom (two translational and one rotational degree of freedom). They are equipped with a gripper which has three rotational degrees of freedom, such that the robots can reach any point in the workspace. The tools can be easily exchanged.

Figure 3: Micro-assembly with microrobots [10]. The glass plate is mounted on an x-y positioning table such that each assembly cell can be brought in the field of view of a microscope. A camera mounted on the microscope forms a sensor system for fine positioning. Coarse positioning of the robots is supervised by a second camera and a laser measurement system. The power of this system lies in the flexibility of the microrobots. Different robot types can operate on the same platform at the same time, each with its own 453

specialisation. One robot can transport parts while another performs assembly, or they can cooperate during an assembly. Furthermore, the robots and their tools can be easily exchanged. 5 MANIPULATION BY PHYSICAL CONTACT Typical problems in assembly are related to the way the part can be picked up, how it can be positioned and how it can be released. 5.1 Sticking effects in microparts handling When parts to be handled are less than one millimetre in size, adhesive forces between gripper and object can be significant compared to gravitational forces. These surface forces can be used in grippers as an adhesive force to pick up the object, but as these forces are almost not controllable, they are more likely to disturb the process rather than to help it. As the gripper approaches, the object can jump off the surface into the gripper, with an orientation depending on the initial charge distribution. When the part is placed at the desired location, it may adhere better to the gripper than to the substrate, preventing accurate placement. These adhesive forces arise primarily from electrostatic attraction, van der Waals forces and surface tension. The balance between these forces depends on the environmental conditions, such as humidity, temperature, surrounding medium, surface condition, material, and relative motion. Tsuchitani et al. [11,12] have studied the surface forces in microstructures. They concluded that the dominant surface force in usual microstructures is (1) the liquid bridge force due to the capillary condensation of the water when the humidity of the atmosphere around the two contacting surfaces is high (over 60% RH); (2) the hydrogen bonding force between water molecules adsorbed on the two surfaces when the humidity is relatively low; and (3) the van der Waals force when adsorbed water molecules on the surface have almost disappeared. Sticking effects are not only problematic for assembly, they also create problems during production of microstructures. For instance, surface machined cantilever beams can stick to the substrate after removal of the sacrificial layer. The surface tension of the rinse liquid is sufficiently strong to pull the suspended cantilever in contact with the substrate, leading to permanent adhesion by van der Waals bonding. Electrostatic force The electrostatic forces arise from charge generation (triboelectrification) or charge transfer during contact. The force per unit area (pressure) for parallel plates is:

p=

σ 2 1 2 εE = s , 2 2ε

(1)

with ε is the permittivity of the dielectric, E the electric field strength and σs the surface charge density. At atmospheric pressure and centimetre-size gaps, the breakdown strength of air (about 30 kV/cm) limits the -5 2 maximum charge density to about 3 x 10 C/m , or peak pressures of about 50 Pa. However, at very small gaps of the order of 1 µm (less than the mean free path of an electron in air), fields of two orders of magnitude higher have been observed. With good insulators such as smooth silica and mica, the charge density can rise up to 2 10 mC/m with pressures in the order of 1 MPa at 1 µm distance [13]. When two materials with different contact potentials are brought in contact, charge flows between them to

equalise this potential. Consider two metal spheres (insulated from their surroundings) brought into contact, then slowly separated. With a contact potential of 0.5 V, 2 the initial charge density will be about 4 mC/m , with field strengths of 5 MV/cm [13]. For small gaps (order 1 nm), electron tunneling and field emission will transfer charge, and for larger gaps (order 1 µm) air breakdown can occur. In principle, using conductive grippers can reduce static charging effects. However, the objects to be handled, such as silicon parts, may be covered with insulators, such as native oxides. On silicon, up to 1 nm of native oxide can build up after several days in air at room temperature. This native oxide is a very good insulator and can withstand a maximum field strength of up to 30 MV/cm [13]. This implies that significant amounts of charge can be stored in the oxide. With the permittivity of silicon (ε = 3.9 ε0), peak pressures are of the order of 100 MPa. When a grounded gripper grasps an initially charged object, charges will be induced in the regions of the dielectric that are not in contact. Surface roughness can prevent charge neutralisation through intimate contact of oppositely charged regions. The residual charges can cause adhesion. Van der Waals force The van der Waals force (sometimes called London’s or dispersion force) is the force that holds together the inert gas crystals and many plastics. It is a much weaker bond than ion and covalent bonds. Many molecules that are too stable to become an integral part, interact with each other through the van der Waals force. Van der Waals forces are due to instantaneous polarisation of atoms and molecules when they are set close. The force between a sphere and a flat gripper can be approximated by [ 14]:

Fvdw =

Hr 6z 2

z