- Email: [email protected]
Some aspects of tribological behaviour at the micro-scale – with particular reference to MEMS and MMAs
Life Cycle Tribology D. Dowson et al. (Editors) © 2005 Elsevier B.V. All rights reserved
77
Some aspects of tribological behaviour at the micro-scale - with particular reference to MEMS and MMAs J. A. Williams and H. R. Le Cambridge University Engineering Department, Trumpington Street, Cambridge CB2 1PZ, UK
Micro-electro-mechanical systems, MEMS, is a rapidly growing interdisciplinary technology dealing with the design and manufacture of miniaturised machines or moving mechanical assemblies (MMAs) with major dimensions at the scale of tens, to perhaps hundreds, of microns. Because they depend on the cube of a representative dimension, component masses and inertias rapidly become small as size decreases whereas surface and tribological effects, which often depend on area, become increasingly important. Although our explanations of macroscopic tribological phenomena often involve individual events occurring at the small scale, when the overall component size is itself miniaturised it may be necessary to re-evaluate some conventional tribological paradigms. This review will look at some of the ways in which for a mechanical engineer small may well be different. 1. INTRODUCTION In December 1959, the physicist Richard Feynman presented a lecture entitled 'There's plenty of room at the bottom' to the American Physical Society at the California Institute of Technology in which he discussed the many challenging possibilities that arise from developing engineering technologies on a very small-scale [Feynman 1992]. Since that time, interest in the applications of miniature and micro-scale machines has grown along with the technologies needed to produce them. The most significant thrust has been to develop design and manufacturing procedures based on fabrication techniques originally established for the production of semi-conductor electronic devices. The advantage of this philosophy being the integration of mechanical functionality (and here are included thermal, fluidic and chemical aspects) with some form of electronic signal processing on the same chip. When produced in sufficient numbers, the economies of scale hold out the promise that such integrated systems will be low cost — comparable to the unit costs of today's semiconductor ICs. The archetypal device of this sort is the microsensor accelerometer used to trigger vehicle airbags (illustrated in Fig 1). The magnitude of the acceleration of the vehicle is detected through its effect on capacitive sensors which form part of the proof mass and the resulting voltage is amplified to a level sufficient to release the trigger of the safety device. Energy has thus been converted from the mechanical to the electrical domain and the device is truly a micro-electro-mechanical system.
Fig. 1 MEMS accelerometer. Motion of the proof mass is detected by changes of capacitance between the elements of the 'combs': courtesy Analog Devices. A spectacular micro-conversion of electrical energy to mechanical is, without doubt, the Texas Instrument Digital Micromirror Device illustrated in Figs 2 and 3. Put at its simplest, this is an array of microscopically small, square mirrors - totalling more than half a million - in an area of about 1.5 square cm. Each mirror, which corresponds to a single pixel on the projected display, can be individually addressed and rotated about one or other of its diagonals through an angle of about 20° at a frequency of more than 100 kHz. What these two examples have in common is their base material - silicon - and the process route by which they have been fabricated. In neither case are they assembled from individual components but both are produced by a sequence of photolithography, chemical etching and deposition of further layers of material: this is known collectively as silicon surface micro-machining [Linder 1999, Elwenspoek 1999].
78
Fig. 2 The Texas instrument Digital Micro-mirror Device consists of more than 5xl05 individually movable mirrors each ca. 20pm square.
metal layer to provide conductive paths. There are up to eight intermediate photolithography stages required for patterning the layers. The sequence for a motor is illustrated in Fig. 5. The perforations visible in the large planar areas of Fig. 1 allow access of etchant to the sacrificial oxide layer removed at the final, release stage of the processing.
Silicon Substrtf «
1 Oxide-1
Silicon Subrtrtf «
Silicon Subatrtf*
2 Poiy-1
Silicon Substr^e
Fig. 3 Electrostatic forces deflect the mirror about one of its diagonals at rates of more than 100 kHz [Bustillo 1998]. Figure 4 is a micrograph of an electrostatic motor fabricated by a similar route. The central rotor spins about a fixed hub driven by appropriately phased voltages applied to the circumferential metallic deposited electrodes on the stator. The central hub is ca. 20 urn in diameter - less than the size of a human hair - and the rotational speed is of the order of 30,000 rpm.
Fig. 4 An electrostatic MEMS "wobble" motor. In each of these devices, the final stage of the fabrication route is to release, by chemical etching, the components which must for functionality articulate relative to both each other and the fixed elements of the design. This is possible, firstly, because both poly-crystalline silicon and silicon oxide can be readily deposited as thin films, and secondly, because the silicon oxide interlayers can be preferentially removed by suitable etchants in either the liquid or vapour phase. The motor in Fig 4 was generated using the socalled MUMPS™ sequence (Multi-User MEMSProcess) which is a seven layer route: an initial silicon nitride isolation layer, two sacrificial oxide oxide layers, three poly-crystalline silicon (usually known as polysilicon or 'poly') layers and a final
Silicon -iut-itfda
3
Silicon Substrii*
4
Silicon Substrria
Fig. 5 A three layer fabrication process. Each deposition stage is preceded by photolithography to generate the mask necessary to give the desired pattern (1) an initial isolation layer of silicon nitride is deposited on the polished Si wafer substrate (2) the first SiO2 layer is deposited, this is followed (3) by deposition of a layer of polysilicon (4) Deposition of a second oxide layer and (5) the second polysilicon layer (6) evaporation of a suitable metal again through a photolithographic mask. The final stage (7) is to etch away the sacrificial oxide layers; the rotor is now free to rotate about the fixed central hub. Several concerns about these designs will occur to a mechanical engineer. In the case of the micromirrors supported on effective flexing beams surely fatigue is going to be a problem - after all, cyclic fatigue is the most commonly encountered mode of failure in structural materials and these flexures may undergo more than 10" cycles of operation every ten hours of use. But they are made of silicon, a brittle, quasi-ceramic material, in which there is no observed dislocation activity at temperatures below 500°C [Muhlstein 2003]. Silicon should not fatigue at temperatures below this and indeed there is no evidence of any hinge failure due to this mechanism. The manufacturers have tested the DMD to more than a trillion, i.e. 1012, cycles and report no evidence of hinge breakages due to fatigue. A further feature evident from Fig. 5 is that the articulating components are cut, as it were, from a single sheet of material by a 'chemical fret-saw'. The saw-cut, which takes the form of an etched
79 trench, has a finite width, typically a few microns, and this becomes a measure of the back-lash between the released components. This is clear in Fig 6(a) which illustrates a pair of meshing gears with a ratchet arrangement on one to prevent reverse rotation: Fig. 6(b) illustrates details of the mesh itself showing the comparatively poor tolerances that are a consequence of the fabrication route. A typical macro-journal bearing will have a clearance ratio of 1 part in 1000, say 20 microns in a bearing 20 mm in diameter. By contrast, in a MEMS bush with a diameter of say 50 um, the diametrical clearance would be about 0.5 um, so ten times less accurate.
electric motors which depend on electromagnetic force generation. One of the 'standard' primemovers in a MEMS machine is an electrostatic comb-motor: an alternating voltage is applied to the interdigitated elements of an electrical capacitor in which one set of fingers or plates is fixed and the other capable of motion by virtue of being supported on some form of compliant structure. Such an arrangement is illustrated in Fig. 7.
• t.
I Fig. 6 (a) MEMS ratchet (b) details of a MEMS gear mesh (c) 'clip' to ensure maintenance of meshing gears: courtesy Sandia National Laboratories, SUMMIT1M Technologies, www.mems.sandia.gov The devices shown in Fig. 6 were produced by the SUMMiT process (Sandia Ultra-planar Multilevel MEMS Technology) which is capable of depositing up to five structural or functional layers. Polysilicon layers in MEMS are thin - typically no more than 3 microns in thickness - so that the planar components themselves have very little stiffness in the out-of-plane direction: to ensure continuity of contact some form of restraint or clip will often be needed as part of the assembly as illustrated in Fig. 6 (c). 2. INFLUENCE OF SCALING If we consider the effect of scaling down the linear dimensions of some macro-scopic device or assembly by a factor of between 100 and 1000, in other words from characteristic dimensions of cm to tens or hundreds of microns, then volume - and by extension mass or inertia - reduces by a factor of at least 106. While it is an oversimplification to say that the MEMS universe is one without inertia, it is true that inertia loads are rarely a significant design constraint. For example, in the DMD illustrated in Fig. 2, the impact force with which the mirror hits the stop as it decelerates is only of the order of a few hundred [M: the resultant stresses are not sufficient to cause any significant mechanical damage. Electromagnetic forces scale principally with volume so that conversion of electrical energy into mechanical energy at the microscale is more likely to involve electrostatic or piezo-electric drives, which scale with area, than small 'conventional'
Fig. 7 SEM micrograph of a poly/SiC lateral resonant MEMS structure. The central beam-like proof mass is supported by two folded-beam ' springs and excited by two electrostatic combmotors [Song et al 2001J. To maximise the efficiency of energy conversion from the electrical to the mechanical domains, such devices are operated near resonance typically at 30 or 40 kHz. Conversion of electrical energy to mechanical by a thermal route can also conveniently be brought about by using a array of thermal bimorphs, one of which is illustrated in Fig. 8. When current is passed from the two anchor points both arms of the device are subjected to ohmic heating; however, because of the very large difference in cross-sectional area the temperature rise in the narrower limb is very much greater than that in the broader. The consequential differential thermal expansion leads to motion of the tip in the direction shown. When the current is removed the bimorph returns elastically to its rest state. At the MEMS scale, the current can be cycled at the rate of kHz.
Direction of Motion
Anchor
Fig. 8 A single thermo-elastic bimorph actuator. It is clear that the assemblies shown in Figs. 7 and 8 produce reciprocating motion. If the end use of the device requires rotation — perhaps at a steady
80 rotational speed - then some form of essentially planar linkage is required to bring about the conversion. A possible design is shown in Fig. 9(a) which allows the kinematic combination of the linear motion from two comb-drives. -kyy -dyfly«t
h
surface tension, manifest as capillary forces or as Laplace pressure differences, important but rarely dominant at the macro-scale, can be very significant, maybe, at the micron scale, to the extent of preventing the desired degree of movement. The presence of a liquid is usually the by-product of the release process but can also form as a result of local humidity and condensation. While capillary forces themselves can be sufficient to cause sticking, an additional problem arises when the liquid starts to evaporate. The resultant volume reduction can produce forces of sufficient magnitude to collapse relatively fragile MEMS structures [Patton 2000, Bhushan 2003].
Fig. 9 (a) Conversion of reciprocating motion into rotation requires some form of planar mechanical linkage with several turning joints [Miller et al 1996] as embodied in a MEMS micro-engine Miller et al 1997]. An embodiment of this principle (albeit for a slightly different geometry) is shown in Fig. 9(b). The pinion is effectively the disc crank of radius r in Fig. 9(a) and the two slender members in the LH corner of Fig. 9(b) are the two connecting rods of Fig. 9(a). If unsteady or incremental rotary motion is acceptable then simpler mechanisms are possible. For example, Fig. 10 shows the way in which an array of thermo-elastic bimorphs can be configured to give a form of MEMS rotary stepper motor. In all these devices there are clearly tribological elements, and thus opportunities and challenges, since all involve turning joints at which friction and wear must be controlled and minimised. The other side of the coin to being freed of the tyranny of mass is the fact that forces associated with area become more important - specifically these include electrostatic, van der Waals, viscous drag and piezoelectric effects. If two clean, similar, conformal surfaces in a MEMS device come into contact they are likely to stick because of the reduction in surface energy that occurs at such a conjunction. 'Stiction' is much more important at the micro-scale than at the macro-scale. This can be a major headache during the final stage of the fabrication process when the moving elements of the assembly must break free of the fixed components. Sometimes they don't and must be encouraged to do so: micro-air jets have been used to overcome initial stiction and enable microengines to run [Gabriel 1990], but in many cases a stuck MEMS device is junk. At the macro-scale we are accustomed to overcoming problems associated with friction by providing a lubricant film. At the micro-scale fluid film lubrication is not an option. The forces of
Fig. 10 Thermal rotary stepper motor. Each 'step' involves (i) engaging the ratchet (ii) unlocking the large gear wheel (Hi) activating the thermal bimorph array to drive the plate from left to right and the gear clockwise, and (iv) engaging the locking plate, disengaging the ratchet plate and allowing the bimorphs to cool and recover elastically to the start position. Control of wear is likewise crucial: in the ratchet devices illustrated in Fig. 10, or in their linear analogues (sometimes called 'scratch motors') loss of key geometry within the drive arrangement will lead to loss of function. The life of the wobble motor shown in Fig. 4 is similarly limited by wear, in this case of the central hub around which the rotor precesses in response to the changing sequence of electrostatic fields. Excessive wear at the centre can lead to contact at the circumference and so loss of function.
Fig. 11 Evidence of wear in a plain journal bearing in a silicon MEMS device [Tanner et al 1999].
81
Figure 11 shows a plain bearing in MEMS device before and after testing. The rotating journal carried a steady load and the resultant loss of silicon material is very clear - some of the finely divided debris can be seen around the edge of the bearing. The assembly of which this was part ran for 158,000 cycles; the shaft speed was in excess of over 100,000 rpm and this damage occurred in just over 90 seconds.
history. Napoleon's watch maker is reputed to have made the remark 'Give me a perfect oil and I will give you a perfect watch'. 1 m/s
1 mm
1 pm/s
length 1 [im
-:.-
3. TRIBOLOGICAL CHALLENGES IN MEMS In one sense thinking about friction, wear and lubrication at the small scale — the micro-scale — is not new to the tribological community. We have become accustomed to explanations of everyday macroscopic behaviour which depend on invoking microscopic or even molecular phenomena. For example, in order to explain the constancy of the coefficient of friction between macroscopic sliding components we reach for a model which requires a large population of small contact areas distributed within the nominal macroscopic geometric footprint of the contact and couple this with some rheological model of a surface film behaviour. Within the MEMS regime, the tribological aims remain reassuringly familiar - to control friction and minimise wear. Although MEMS devices are small they are micro- rather than nano. A 10 micron cube of silicon contains in excess of 1013 atoms so that there are unlikely to be dramatic changes in mechanical properties from tabulated or familiar values. Having said that, it is not wise to assume that there are no changes - a plot of characteristic length versus characteristic event time is instructive - Fig. 12. On log scales contours of constant sliding speed plot as straight lines with 45° slope. Engineering components will cover a length of perhaps from something less than a millimetre to metres and sliding speeds of less than Imm/s to perhaps 10 m/s. In MEMS devices the length scale of interest stretches from sub-millimetre to nanometres. Although the rotational speed in any MEMS motor is likely to be high, the sliding speed at the bearing surfaces is of the same order as in macroscopic devices. The diagram illustrates the fact that the particular challenge offered by MEMS design is that the technology is really just at the interface between two areas of professional expertise, viz. those of engineers and material scientists. This figure also illustrates how far from the area of MEMS operation are conditions, specifically sliding speed, within the Atomic Force Microscope. The devices illustrated in Figs. 1-11 represent a new order of smallness but the importance of friction and lubrication in fine machinery has a long
MICRO-
^ \
" " ' ' " • " "STRUCTURE
ATOMIC \
.'
STRUCTURE Jyimmic /
fn!
IUl
1,/s
/
/
•
A
MKMS.MMA->
"
/ 1 ms time
Fig. 12 Characteristic scales and contact times for various components and devices: typical operation of MEMS occurs close to the intersection of the areas associated with engineering and materials science. Tribological phenomena at this size range can have significant economic consequences. Moving coil instruments relied for their accuracy on low friction, low-wear pivots, often in the form of jewel bearings. During the 1930s the National Physical Laboratory carried out an extensive programme of experimental work into the design and operation of such bearings with particular reference to those found in electricity meters [Scott 1936]. One experimental run - to measure steadystate friction and wear performance - lasted more than 3.5 years. The results of this study were reported at the renowned General Discussion on Lubrication in 1937 [Scott 1937]. One of the first ideas that neophytes to tribology are made aware of is the notion that flat surfaces are not flat and that the tribological action happens at the tips of interacting asperities. This is (presumably) true at the MEMS scale and Greenwood and Williamson style models of surface roughness are much in evidence in the MEMS literature: there are, however, some additional difficulties in moving from the macro-world to the micro because the characteristic lengths that describe the nominal area of contact and the topography of the surfaces are scaled by different factors: this is compounded by the fact that the roughness values can vary from one poly layer to another and are different on the side walls of the etched trenches from values on the planar surfaces [Phinney 2004]. In addition, at this scale, surface energy terms may not be insignificant so that the Hertzian analysis must be extended to include their effect - for example, through either the JohnsonKendall-Roberts [1971] or Derjaguin-MullerToporov [1975] approaches as is appropriate.
82 3.1 Controlling friction in MEMS - Self Assembled Monolayers MEMS surfaces may come into contact unintentionally through unforeseen acceleration or electrostatic effects, or intentionally, when surface impact normally or shear past one another. If adhesive attraction exceeds the restoring forces, adhesion is permanent and the device has failed due to 'in-use stiction'. Approaches to reduce this can be classified under two headings, viz. physical and chemical modifications of the basic poly-silicon surface. In the physical approach the surfaces are roughened to reduce the effective contact area usually be selective etching, and effective reductions in surface energy as large as factors of 25 have been reported [Yee 1996]. In the chemical approach, the surface chemistry is altered to reduce the intrinsic surface energy and thus the propensity to stick. One technique is to treat the silicon surface with hydrofluoric acid which etches away the hydrophilic native surface oxide and terminates the Si with hydrogen. Since the Si-H bond is non-polar the resulting surface is hydrophobic and so has reduced in-use stiction. While this is attractive for devices encapsulated in inert atmospheres, it is known that the hydrogen terminated bonds are gradually re-oxidised in air so that the treatment is not permanent - although the effect may last several hundred hours which may be sufficient to provide commercial viability to the product. A more effective chemical surface modification involves the application of a mono-molecular film to the micromachine surface. Such Self-Assembled Monolayers, often deposited from solution, when correctly integrated into the manufacturing route, can provide a number of benefits. Firstly, they can effectively eliminate stiction due to meniscus effects during the release stage in wet-etch fabrication (by increasing the contact angle to more than 90° so rendering the surface hydrophobic rather than hydrophilic). Secondly, they can reduce the likelihood of the dry surfaces sticking (by a reduction in effective surface energy) and finally they may act as a boundary lubricant during use so reducing frictional losses and wear volumes. Table I Effect of SAMs on SiO2 surfaces Surface contact angle work of adhesion COF with water against Si degrees mJ/m2 SiO2 <30 20 -l(dry) OTS 110 0.012 0.07 Successful SAMs are robust organic molecules that are chemically adsorbed onto the Si or SiC>2 surface. Characteristically, they have a hydrophilic (polar) head that is bonded to the surface and a long,
hydrophobic tail that extends outwards. Octodecyltrichlorosilane (OTS) CigHsvSiClj is typical and has been intensively studied - its beneficial effects on SiO2 are indicated in Table I although it suffers from the potentially serious defect that the first step of the reaction sequence for binding the molecule to the surface involves the hydrolysis of the original Si-CI bonds thus liberating HC1 which could clearly be detrimental to any exposed metallic surfaces within the MEMS package. Perfluoropolyethers (PFPEs) are routinely used as relatively thick films as lubricants for magnetic storage media - both hard discs and flexible tapes. They have low surface tensions so reduce stiction, low vapour pressures and thus suffer very low evaporative losses and are oxidatively stable. The nature of the bond between the lubricant and the native silicon oxide surface depends on the end group of the PFPE - it may be non-polar (CF3) and therefore relatively weak as in Z-15- or polar (-OH) as in Z-DOL and can be thermally bonded by heating to 150°C for ca. 30 mins. The effect on friction - if not on long term wear - can be seen in Fig. 13. QKirictau of Diction 0.1 Si
0
tS
SO
75
100 125 Numlvr of cycles
(a) (b) Fig. 13 (a) Schematic of AFM tip sliding on Si(100) surface with native oxide (upper), non-polar Z-15 (mid) and polar Z-DOL (lower), (b) Coefficient of friction versus number of sliding cycles [after Liu and Bhushan 2003]. Much of the work into the frictional behaviour of SAMs and PFPEs (for example that illustrated in Fig. 13) has been carried out using an AFM as a 'scanning probe' device: the tip of the AFM probe is extremely fine so that even at very low loads the contact pressure is very much greater than any anticipated 'asperity pressure' in practice and (as has been noted in Fig. 11) the sliding speed is very low. This can be seen in a ' P V map as suggested by Dugger [2001, 2003] and illustrated in Fig. 14: here P is the likely maximum asperity pressure and V the imposed sliding speed. The Surface Force Apparatus [Homola 1989] can generate more
83 realistic sliding speeds but, as it relies on the contact between two relatively fragile mica surfaces, the load or pressure is limited. To some extent the first of these objections can be overcome, albeit with experimental difficulty, by sticking a small glass bead on the AFM tip to increase dramatically the effective radius of curvature; however, a number of commercial instruments are now becoming available which explore the more realistic region of the map designated 'SMM'. With MEMS, as in other situations where success depends on a degree of marginal lubrication, the boundary film must be sufficiently robust to survive for as many cycles of operation as is required - or else, some means of replenishment be provided - this was a significant concern in the design of the Tl Digital Micro-mirror Device. -re-
1000 800
I %
S o.
600
atomistic simulation
scanning probes surface forces apparatus
Fig. 15 Structure of (a) diamond, and (b) graphite, (c) The ternary phase diagram for amorphous carbon-hydrogen alloys.
~(iengine speed record
gear driving a load
400 200
"0
phases. Deposition methods have been developed to produce a-Cs with increasing sp3 bonding including the region of tetrahedral amorphous carbon designated ta-C. However, there is another important variable, viz the amount of hydrogen contained within the structure. This can vary from less than a few atomic percent to more than 50% and the extent of hydrogenation can be shown on a ternary phase diagram as in Fig. 15(c). Some coating methods, such as plasma enhanced chemical vapour deposition, are able to extend into the interior of the triangle producing hydrogenated amorphous carbon or a-C:H.
50
100 150 200 250
800
1000
Velocity, mm/s
Fig. 14 Experimental techniques, atomistic simulations and operating regimes of surface micromachines in pressure-velocity space. SMM refers to the Sandia sidewall tribometer referred to in the text [after Bugger 2001J. 3.2 Diamond like Carbon Films Diamond like Carbon (DLC) is a term used to describe a family of carbon based materials of whom many have attractive tribological properties - such as low friction coefficients (when slid dry against themselves or a variety of other materials) low wear rates, high values of elastic modulus, chemical inertness and good stability. Carbon can form a great variety of crystalline and disordered structures because it is able to exist in three hybridisations known as sp3, sp2 and sp1. In the sp3 configuration, as in natural diamond, a carbon atom's four valence electrons are assigned tetrahedrally as shown in Fig. 15(a): in the sp2, or graphitic, configuration, three of the four valence electrons lie in a plane, to which the final bond, now much weaker, is normal, Fig. 15(b). The so-called amorphous carbons, conventionally designated a-C, contain a mixture of sp2 and sp
Both types of film can, by appropriate choice of deposition conditions, be extremely uniform in thickness (with an intrinsic 'roughness' measured in tenths of nanometres) and thus, since they are only a few hundred nanometres in thickness, they retain the topography of the substrate on which they are laid down. Their mechanical properties can also be varied but in general ta-C has higher values of modulus and hardness than hydrogenated a-C:H as indicated in Table II. Table II Properties of DLC films compared to natural diamond [after Robertson 2002 and Grill 19971 ta-C a-C:H diamond H(at%) 0 30 sp fraction >0.8 <0.5 density (kg mf3) 3260 2350 3515 757 300 E (GPa) 1145 >20 <15 45 hardness (GPa) <1 residual stress (GPa) >6 These data also indicate one of the major problems with ta-C films, viz. the very high level of residual compressive stresses with which they are left. These can be annealed out at temperatures above ca. 400°C but this is unlikely to be feasible in a fabricated MEMS device. At the level of GPa there is a danger of the film delaminating from the substrate or distorting the components within the MEMS. A possible way forward is to create a nanostructured DLC layer by depositing a sequence of alternate harder ta-C and softer aC:H layers: in this
84 way it may be possible to build up a sufficient thickness for DLC to be used as a structural material rather than just a coating. Although the contact surfaces of such carbon films are chemically inert and so would not be expected to enter into major chemical activity with their environment they can still interact through physical adsorption and this mechanism can introduce significant fluctuations in friction and wear. For hydrogenated a-C:H the value of the COF when sliding against itself depends strongly on the relative humidity, see Fig. 16. Values below 0.05 are characteristic of dry air (or vacuum) but these rise by a factor three at more usual ambient values. The value of the COF also appears to be influenced by the ratio of H:C in the precursor gas used in the deposition. Tetrahedral ta-C and ta-C:H behave differently. They have higher COFs in vacuo or dry air but this seems to reduce - or at least not increase - in the presence of humidity.
Table III Super-lubncious DLC films [after Erdemir 2000] Carbon film COF grown in mm'N m 100%CH4 0.015 9.0x10" 25%H 2 +10%CH 4 0.01 7.36xlO'9 50%H 2 +10%CH 4 0.004 1.23xlO~8 90%H 2 +10%CH 4 0.004 2.8xlO"10 100%C2H2 0.27 7.5xl0~7 hydrogen free DLC 0.65 2.8x10"7 uncoated H13 0.8 4.6xlO~6 Microscopic examination of the DLC coated spheres and flats revealed very shallow wear racks and scars even after many hours of running: a longterm test of more 32 days was only terminated when the drive motor of the rig burnt out - the friction coefficient was consistently less than 0.005. The authors suggest that the remarkably low values of COF generated in this study are associated with the increasingly inert nature of the DLC surface with increasing availability of hydrogen during the deposition process. It is suggested that hydrogen passivated the dangling surface bonds that can otherwise lead to increases in friction and wear. 4. MEASURING FRICTION AND WEAR IN MEMS DEVICES
0.3 1 3 Relative humidity (%)
Fig. 16 Variation of friction coefficient with relative humidity for a-C:H and ta-C, after Robertson [2002], Enke [1981] and Voevodin [1996] When discussing the tribological properties of DLC, as with anything else, it is important to take account of the nature of the counterface. For a number of different counterface materials (including steel, silicon nitride and sapphire) long-term stability appears dependent on the formation of a carbon-rich transfer layer derived from the DLC but with a distinctive morphology of its own often consisting of fine graphitic nano-particles (<5 nm) within a distorted diamond-like structure. If both the extrinsic or environmental factors and the intrinsic properties of the film and the underlying substrates can be optimised then it is possible to generate extraordinarily low values of both COF and specific wear rates - see Table III. In these experiments the DLC films were deposited on extremely smooth (< 100 nm Ra) tool steel surfaces and tested in a ball on flat tribometer where the peak hertzian pressure was 1.04 GPa.
It is generally considered poor practice to run likeon-like surfaces in any tribological contact and so it is not perhaps surprising that silicon-on-silicon is no exception to this. The qualitative impression, for example the evidence of Fig. 11, is that wear rates in plain bearings in which two poly components run against one another are unacceptably high. If we are to allow for wear at the design stage we need some knowledge of the Archard wear constant - together, of course, with the assumption that this really is a 'constant'. Literature values of Kw for bare poly-onpoly at the macro-scale vary between 10" m m W 1 and 10~6 m m V m 1 [Gardos 1998]. The experimental determination of this number at the appropriate scale is not-straight-forward as we need to control load and sliding distance and measure wear volume loss within reasonable tolerance limits. It is difficult to extract data of this sort from experience with operational MEMS motors or MMAs because, in many cases, even if the sliding distance is known and wear volumes estimated from micrographs, the contact load is not measured or known with sufficient accuracy [Mehregany 1992, Williams 2001]. Several research groups are rising to the challenge to build a MEMS tribometer in which the necessary parameters are controlled and the desired variables measured. Figure 17 shows an example of
85 one such device, built by the Sandia Labs, which generates reciprocating motion, with stroke A, generated by flexing the poly-Silicon cantilever by means of electrostatic effects [deBoer 1998]. Actuation is performed under an interferemetric microscope so that deflections - and thus loads - can be assessed. An immediate problem with this arrangement is that A is small, only about 30 nm, so that it is not really feasible to assess wear quantitatively, but the nominal contact pressure can be varied from less than 10 kPa to about 25 MPa. To get useful work out of a functional MEMS or MMA will need nominal contact pressures of the order of a MPa. Observed static COFs between poly surfaces were between 2 and as much as 8: under sliding conditions this reduced very considerably to ca. 0.3 for silicon on silicon, and 0.16 for Si surfaces carrying a SAM.
suspended beam into contact with a fixed or anchored post under a known load, and the other to oscillate the beam against the post. The contact geometry is thus equivalent to a nominal line contact. Since the device itself is a MEMS the surface morphology and chemistry duplicate precisely those found in more complicated systems having contacting surfaces. The device operates in a PV range representative of other surface micromachined devices, see Fig. 14. A more recent device designed to characterize friction between micromachined surfaces is an 'nanotractor' device also developed at Sandia and shown in Fig. 19 [deBoer 2004]. It consists of a set of rails on the bottom of a moving structure that make contact with a planar track so that by repeating the sequence of electrostatically pulling one clamp down, deflecting the actuation plate, pulling the other clamp down and then releasing the first clamp and the actuation plate, the device can be walked out against the restoring force of the suspension springs and "load cell" beam. Static friction is determined by gradually releasing the clamping voltage until the device slips.
spring restoring force, lys
Fig. \1 (a) The Sandia 'Lab-on-a-chip' tribology tester in which the electrostatically driven flexure of the beam can be quantified by viewing by optical interference (b) [after deBoer et al 1998]. A similar device, which includes an additional link to extend the stroke is being used by a group at Dayton Research Institute [Patton 2000]. The device currently being used by the group at Berkeley is similar in some ways to the accelerometer illustrated in Fig.l. A planar shuttle, which carries a number of raised pimples on its lower surface, is actively reciprocated by a set of comb motors: simultaneously shuttle and substrate are pulled together by a second electrostatic arrangement so establishing a controlled load on the tips of the pimples. Nominal pressures are in the range 0.1 to 2.5 MPa and experiments can be carried out in air and in vacuum. 4.1 Sandia 'Side-wall' tribometer Sandia labs have built a further generation of devices to look specifically at sliding on the sidewalls of the chemically etched trenches on the basis that these are often the surfaces of interest in a functioning MEMS and are likely to have different topographies from those of the planar wafer surfaces. The device, the 'sidewall microtribometer' is shown in Fig. 18. It is driven by two electrostatic comb motors, one is used to pull a
^
ink
|
comb drive force. F c -Mo
Fig. 18 The Sandia 'side-wall' tribometer. Electrostatic actuators pull the suspended beam into contact with the anchored post while a second set oscillates the beam back and forth (Dugger 2003b). Dynamic friction is determined by releasing the clamping voltage and allowing the device to oscillate while measuring the displacement versus time. A damped oscillator model is used to determine the air damping and dynamic friction coefficients required to fit the displacement data. Parallel plate capacitors having relatively large area,
86 and the ability to tailor the size of the rails, enable investigation of a wide range of contact pressures.
Fig 19 The Sandia 'nanotractor device [deBoer 2004] 4.2 CUED micro-tribometer The Sandia devices illustrated above make use of the multilevel SUMMiT™ process route alluded to in §1. We have recently commissioned a test-rig in which the dimensions of the bearing under test approach the micro-scale and in which it is the only sliding contact downstream of any load-measuring element - but which is not itself a MEMS device. The test surfaces, which can be of silicon, form a thrust bearing between the appropriately profiled upper end of a vertical rotating shaft and a matching 'dimple' in a small circular disc which sits within a symmetrical 'top hat' supported by the shaft as illustrated in Fig. 20. The mass of the top-hat provides the load on the test bearing and its rotational speed is monitored by a non-contact probe.
the test thrust bearing. Initially this has been set up as a sphere of radius ca. 1 mm within a spherical dimple with a slightly greater radius. With different top-hats and shaft tips it is possible, by varying the nature of the bearing geometry, to generate effective maximum contact pressures from a few MPa to over 1 GPa. In principle, measuring the friction in this device is very straightforward. If the shaft is accelerated up to some speed and then the motor stopped the tophat decelerates at a constant rate given by the frictional torque divided by its polar moment of inertia - which is known. The geometry of the contact (ball in cup, ball in cone etc) provides an effective radius at which friction acts and thus the coefficient of friction between the two surfaces concerned can be calculated. To generate wear requires some slip between the driving and the driven surfaces of the bearing and this can be achieved by cycling the motor speed and counting the number of 'slip revolutions' that occur during each phase of acceleration and deceleration. However, a neater method of providing some retarding torque on the top-hat so that a steady-state is reached with the shaft rotating at one speed and the specimen moving relative to it at some fixed slip speed is to use the rim of the hat as the disc in an eddy-current disc brake. The specimens can be cut from a regular Si wafer and given the same sequence of surface treatments as a MEMS device so that the rig is appropriate for the evaluation of several of the suggested routes to tribological success such as laser texturing or the combination of hydrophilic and hydrophobic surfaces in a so-called half-wetted bearing [Spikes 2003]. 5. CONCLUSIONS
Fig. 20 CUED Micro-tribometer capable of generating contact conditions very similar to those within MEMS devices. The apparatus sits within a bell jar which can be evacuated so that the only torque acting to accelerate or decelerate the top-hat arises from friction within
• Micro-systems are rich with issues concerned with the properties, processing and mechanics of materials. Much of current MEMS technology is based on silicon and although this material has some attractive properties - particularly its fatigue resistance - it is less attractive tribologically. Alternative candidate materials must be compatible with MEMS fabrication routes. • Macroscopic machines are often limited by inertia effects: MEMS and MMAs are much more commonly limited by surface forces and the effects stiction/friction and wear. For MEMS, tribology is an important enabling technology. • Sliding speeds in MEMS are by macro standards comparatively modest - i.e. less than 1 ms~'. Sliding distances, in rotating bearings can be of order of hundreds of km if the device is likely to be in continuous use. • For 'reasonable' mechanical performance, nominal bearing pressures must be of the order of
87 MegaPascals and for acceptable life, specific wear rates need to be no more than ca. 1(T8 mm31ST1 m~'. • Within MEMS conventional liquid lubrication is impossible because of meniscus force effects. Self Assembled Monolayers applied to sliding surfaces may provide acceptable combinations of low friction and wear though robustness and replenishment may be issues of concern. • In dry running MEMS and MMAs, Diamond-LikeCarbon has some very attractive tribological features but some deposition routes can leave the film with unacceptably high values of residual stress. • Modelling the behaviour of small-scale contacts is challenging, not only because of the inclusion of surface energy effects but also because of the different factors by which component dimensions and surface topography may change as the scale of the device reduces. 6. ACKNOWLEDEGEMENTS Our thanks are due to Professor Victor Bright of the University of Colorado for Fig. 10 and to Dr Mike Dugger of the Sandia Labs for details of their micro- and nano-tribology facilities.
REFERENCES Bhushan,B (2003) J. Vac. Sci. Technol. B21 2262 Bustillo J M, Muller R S (1998) Trans IEEE 86 1552 Corwin A D, Walraven J A and Redmond J M (2004) J Mems 13 63 deBoer M P, Redmond J M and Michalske T A (1998) SPIE Proc. 3512 Santa Clara deBoer M P, Luck D L, Ashurst W R, Maboudian R, Corwin A D, Walraven J A and J M Redmond J M (2004) J Mems 13 63 Derjaguin B V, Muller V M and Toporov Y P (1975) J. Colloid Interface Sci. 53 314 Douglass TI White Paper: http://www.plusamerica.com/papers.html Dugger M T (2003) Quantification of friction in micro-system contacts in Nanotribology Critical Assessement and Research Needs ed. S M Hsu and ZC Ying, Kluwer, Boston Dugger M T (2003b) Tribology in Microsystems in TRIMS 2003: Microtribology Neuchatel, Switzerland. Elwenspoek M and Jansen H (1999) Silicon Micromachining CUP Encke K (1981) Thin Solids Films 80 227 Erdemir A, Eyrilmaz O L, Nilufer I B and Fenske G R (2000) Surface Coatings and Technol. 133134 448 Feynman R P(1992) reprinted in ASME J. Microelectromechanical Systems 1 60 Gabriel K J, Behi F, Mahadevan R and Mehegrany M (1990), Sensors and Actuators A21-A23 184
Gardos, M (1998) in Tribology issues and Opportunities in MEMS, ed B Bhushan, Kluwer, Dordrecht. Grill A (1997)Surface Coatings and Technol. 94-95 507 Homola A M, Israelachvili J N, Gee M L and McGuiggan P M (1989) Trans ASME J of Tribology 111 675 iMEMS® Integrated Micro-Electro Mechanical Systems, Analog Devices http://www.analog.com Johnson K L, Kendall K and Roberts A D (1971) Proc. Roy. Soc. A324 301 Linder C, Parette L, Gretillat M-A, Jaecklin V P and de Rooij N F (1999) J. Micromech. Microengng 2 122 Liu H and Bhushan B (2003) Ultramicroscopy 97 321 Mehregany M and Senturia S D (1992) IEEE Trans Electron Devices 39 1136 Miller S L and LaVigne G (1997) Proc SPIE Smart Electronics and MEMS 3224 24 Miller S L, Sniegowski J J, LaVigne G and McWhorter P J (1996) Proc SPIE Smart Electronics and MEMS 2722 197 Muhlstein C L, Stach E A and Ritchie R O (2003) Engng. Integrity 14 4 Patton, S T, Cowan W D, Eapen K C and Zabinski J S (2000) Trib Lett 9 199 Robertson J (2002) Materials Science and Engineering Reports, R37 129 Scott V (1936) National Physical Laboratory, Collected Researches 24 1 Scott V (1937) Proc IMechE General discussion on Lubricants and Lubrication, London 2145 Song X, Rajgolpal S, Melzak J M, Zorman CA and Mehegrany M (2001) Mater. Sci. Forum 389-393 755 Spikes H A (2003) Proc. Instn Mech. Engs 217J 1 Tanner D M, Miller W M, Peterson K A, Dugger M T, W P Eaton, Irwin L W, Senft D C, Smith N, Tangyunyong F P and Miller S L (1999) Microelectronics Reliability 39 401 Voevodin, A A, Phelps A W, Zabinski J S and Donley M S (1996) Diamond Rel Mater. 5 1264 Williams J A (2001)Wear 251 965 Yee Y, Chun K and Lee J D (1996) Actuators A52 145
77
Some aspects of tribological behaviour at the micro-scale - with particular reference to MEMS and MMAs J. A. Williams and H. R. Le Cambridge University Engineering Department, Trumpington Street, Cambridge CB2 1PZ, UK
Micro-electro-mechanical systems, MEMS, is a rapidly growing interdisciplinary technology dealing with the design and manufacture of miniaturised machines or moving mechanical assemblies (MMAs) with major dimensions at the scale of tens, to perhaps hundreds, of microns. Because they depend on the cube of a representative dimension, component masses and inertias rapidly become small as size decreases whereas surface and tribological effects, which often depend on area, become increasingly important. Although our explanations of macroscopic tribological phenomena often involve individual events occurring at the small scale, when the overall component size is itself miniaturised it may be necessary to re-evaluate some conventional tribological paradigms. This review will look at some of the ways in which for a mechanical engineer small may well be different. 1. INTRODUCTION In December 1959, the physicist Richard Feynman presented a lecture entitled 'There's plenty of room at the bottom' to the American Physical Society at the California Institute of Technology in which he discussed the many challenging possibilities that arise from developing engineering technologies on a very small-scale [Feynman 1992]. Since that time, interest in the applications of miniature and micro-scale machines has grown along with the technologies needed to produce them. The most significant thrust has been to develop design and manufacturing procedures based on fabrication techniques originally established for the production of semi-conductor electronic devices. The advantage of this philosophy being the integration of mechanical functionality (and here are included thermal, fluidic and chemical aspects) with some form of electronic signal processing on the same chip. When produced in sufficient numbers, the economies of scale hold out the promise that such integrated systems will be low cost — comparable to the unit costs of today's semiconductor ICs. The archetypal device of this sort is the microsensor accelerometer used to trigger vehicle airbags (illustrated in Fig 1). The magnitude of the acceleration of the vehicle is detected through its effect on capacitive sensors which form part of the proof mass and the resulting voltage is amplified to a level sufficient to release the trigger of the safety device. Energy has thus been converted from the mechanical to the electrical domain and the device is truly a micro-electro-mechanical system.
Fig. 1 MEMS accelerometer. Motion of the proof mass is detected by changes of capacitance between the elements of the 'combs': courtesy Analog Devices. A spectacular micro-conversion of electrical energy to mechanical is, without doubt, the Texas Instrument Digital Micromirror Device illustrated in Figs 2 and 3. Put at its simplest, this is an array of microscopically small, square mirrors - totalling more than half a million - in an area of about 1.5 square cm. Each mirror, which corresponds to a single pixel on the projected display, can be individually addressed and rotated about one or other of its diagonals through an angle of about 20° at a frequency of more than 100 kHz. What these two examples have in common is their base material - silicon - and the process route by which they have been fabricated. In neither case are they assembled from individual components but both are produced by a sequence of photolithography, chemical etching and deposition of further layers of material: this is known collectively as silicon surface micro-machining [Linder 1999, Elwenspoek 1999].
78
Fig. 2 The Texas instrument Digital Micro-mirror Device consists of more than 5xl05 individually movable mirrors each ca. 20pm square.
metal layer to provide conductive paths. There are up to eight intermediate photolithography stages required for patterning the layers. The sequence for a motor is illustrated in Fig. 5. The perforations visible in the large planar areas of Fig. 1 allow access of etchant to the sacrificial oxide layer removed at the final, release stage of the processing.
Silicon Substrtf «
1 Oxide-1
Silicon Subrtrtf «
Silicon Subatrtf*
2 Poiy-1
Silicon Substr^e
Fig. 3 Electrostatic forces deflect the mirror about one of its diagonals at rates of more than 100 kHz [Bustillo 1998]. Figure 4 is a micrograph of an electrostatic motor fabricated by a similar route. The central rotor spins about a fixed hub driven by appropriately phased voltages applied to the circumferential metallic deposited electrodes on the stator. The central hub is ca. 20 urn in diameter - less than the size of a human hair - and the rotational speed is of the order of 30,000 rpm.
Fig. 4 An electrostatic MEMS "wobble" motor. In each of these devices, the final stage of the fabrication route is to release, by chemical etching, the components which must for functionality articulate relative to both each other and the fixed elements of the design. This is possible, firstly, because both poly-crystalline silicon and silicon oxide can be readily deposited as thin films, and secondly, because the silicon oxide interlayers can be preferentially removed by suitable etchants in either the liquid or vapour phase. The motor in Fig 4 was generated using the socalled MUMPS™ sequence (Multi-User MEMSProcess) which is a seven layer route: an initial silicon nitride isolation layer, two sacrificial oxide oxide layers, three poly-crystalline silicon (usually known as polysilicon or 'poly') layers and a final
Silicon -iut-itfda
3
Silicon Substrii*
4
Silicon Substrria
Fig. 5 A three layer fabrication process. Each deposition stage is preceded by photolithography to generate the mask necessary to give the desired pattern (1) an initial isolation layer of silicon nitride is deposited on the polished Si wafer substrate (2) the first SiO2 layer is deposited, this is followed (3) by deposition of a layer of polysilicon (4) Deposition of a second oxide layer and (5) the second polysilicon layer (6) evaporation of a suitable metal again through a photolithographic mask. The final stage (7) is to etch away the sacrificial oxide layers; the rotor is now free to rotate about the fixed central hub. Several concerns about these designs will occur to a mechanical engineer. In the case of the micromirrors supported on effective flexing beams surely fatigue is going to be a problem - after all, cyclic fatigue is the most commonly encountered mode of failure in structural materials and these flexures may undergo more than 10" cycles of operation every ten hours of use. But they are made of silicon, a brittle, quasi-ceramic material, in which there is no observed dislocation activity at temperatures below 500°C [Muhlstein 2003]. Silicon should not fatigue at temperatures below this and indeed there is no evidence of any hinge failure due to this mechanism. The manufacturers have tested the DMD to more than a trillion, i.e. 1012, cycles and report no evidence of hinge breakages due to fatigue. A further feature evident from Fig. 5 is that the articulating components are cut, as it were, from a single sheet of material by a 'chemical fret-saw'. The saw-cut, which takes the form of an etched
79 trench, has a finite width, typically a few microns, and this becomes a measure of the back-lash between the released components. This is clear in Fig 6(a) which illustrates a pair of meshing gears with a ratchet arrangement on one to prevent reverse rotation: Fig. 6(b) illustrates details of the mesh itself showing the comparatively poor tolerances that are a consequence of the fabrication route. A typical macro-journal bearing will have a clearance ratio of 1 part in 1000, say 20 microns in a bearing 20 mm in diameter. By contrast, in a MEMS bush with a diameter of say 50 um, the diametrical clearance would be about 0.5 um, so ten times less accurate.
electric motors which depend on electromagnetic force generation. One of the 'standard' primemovers in a MEMS machine is an electrostatic comb-motor: an alternating voltage is applied to the interdigitated elements of an electrical capacitor in which one set of fingers or plates is fixed and the other capable of motion by virtue of being supported on some form of compliant structure. Such an arrangement is illustrated in Fig. 7.
• t.
I Fig. 6 (a) MEMS ratchet (b) details of a MEMS gear mesh (c) 'clip' to ensure maintenance of meshing gears: courtesy Sandia National Laboratories, SUMMIT1M Technologies, www.mems.sandia.gov The devices shown in Fig. 6 were produced by the SUMMiT process (Sandia Ultra-planar Multilevel MEMS Technology) which is capable of depositing up to five structural or functional layers. Polysilicon layers in MEMS are thin - typically no more than 3 microns in thickness - so that the planar components themselves have very little stiffness in the out-of-plane direction: to ensure continuity of contact some form of restraint or clip will often be needed as part of the assembly as illustrated in Fig. 6 (c). 2. INFLUENCE OF SCALING If we consider the effect of scaling down the linear dimensions of some macro-scopic device or assembly by a factor of between 100 and 1000, in other words from characteristic dimensions of cm to tens or hundreds of microns, then volume - and by extension mass or inertia - reduces by a factor of at least 106. While it is an oversimplification to say that the MEMS universe is one without inertia, it is true that inertia loads are rarely a significant design constraint. For example, in the DMD illustrated in Fig. 2, the impact force with which the mirror hits the stop as it decelerates is only of the order of a few hundred [M: the resultant stresses are not sufficient to cause any significant mechanical damage. Electromagnetic forces scale principally with volume so that conversion of electrical energy into mechanical energy at the microscale is more likely to involve electrostatic or piezo-electric drives, which scale with area, than small 'conventional'
Fig. 7 SEM micrograph of a poly/SiC lateral resonant MEMS structure. The central beam-like proof mass is supported by two folded-beam ' springs and excited by two electrostatic combmotors [Song et al 2001J. To maximise the efficiency of energy conversion from the electrical to the mechanical domains, such devices are operated near resonance typically at 30 or 40 kHz. Conversion of electrical energy to mechanical by a thermal route can also conveniently be brought about by using a array of thermal bimorphs, one of which is illustrated in Fig. 8. When current is passed from the two anchor points both arms of the device are subjected to ohmic heating; however, because of the very large difference in cross-sectional area the temperature rise in the narrower limb is very much greater than that in the broader. The consequential differential thermal expansion leads to motion of the tip in the direction shown. When the current is removed the bimorph returns elastically to its rest state. At the MEMS scale, the current can be cycled at the rate of kHz.
Direction of Motion
Anchor
Fig. 8 A single thermo-elastic bimorph actuator. It is clear that the assemblies shown in Figs. 7 and 8 produce reciprocating motion. If the end use of the device requires rotation — perhaps at a steady
80 rotational speed - then some form of essentially planar linkage is required to bring about the conversion. A possible design is shown in Fig. 9(a) which allows the kinematic combination of the linear motion from two comb-drives. -kyy -dyfly«t
h
surface tension, manifest as capillary forces or as Laplace pressure differences, important but rarely dominant at the macro-scale, can be very significant, maybe, at the micron scale, to the extent of preventing the desired degree of movement. The presence of a liquid is usually the by-product of the release process but can also form as a result of local humidity and condensation. While capillary forces themselves can be sufficient to cause sticking, an additional problem arises when the liquid starts to evaporate. The resultant volume reduction can produce forces of sufficient magnitude to collapse relatively fragile MEMS structures [Patton 2000, Bhushan 2003].
Fig. 9 (a) Conversion of reciprocating motion into rotation requires some form of planar mechanical linkage with several turning joints [Miller et al 1996] as embodied in a MEMS micro-engine Miller et al 1997]. An embodiment of this principle (albeit for a slightly different geometry) is shown in Fig. 9(b). The pinion is effectively the disc crank of radius r in Fig. 9(a) and the two slender members in the LH corner of Fig. 9(b) are the two connecting rods of Fig. 9(a). If unsteady or incremental rotary motion is acceptable then simpler mechanisms are possible. For example, Fig. 10 shows the way in which an array of thermo-elastic bimorphs can be configured to give a form of MEMS rotary stepper motor. In all these devices there are clearly tribological elements, and thus opportunities and challenges, since all involve turning joints at which friction and wear must be controlled and minimised. The other side of the coin to being freed of the tyranny of mass is the fact that forces associated with area become more important - specifically these include electrostatic, van der Waals, viscous drag and piezoelectric effects. If two clean, similar, conformal surfaces in a MEMS device come into contact they are likely to stick because of the reduction in surface energy that occurs at such a conjunction. 'Stiction' is much more important at the micro-scale than at the macro-scale. This can be a major headache during the final stage of the fabrication process when the moving elements of the assembly must break free of the fixed components. Sometimes they don't and must be encouraged to do so: micro-air jets have been used to overcome initial stiction and enable microengines to run [Gabriel 1990], but in many cases a stuck MEMS device is junk. At the macro-scale we are accustomed to overcoming problems associated with friction by providing a lubricant film. At the micro-scale fluid film lubrication is not an option. The forces of
Fig. 10 Thermal rotary stepper motor. Each 'step' involves (i) engaging the ratchet (ii) unlocking the large gear wheel (Hi) activating the thermal bimorph array to drive the plate from left to right and the gear clockwise, and (iv) engaging the locking plate, disengaging the ratchet plate and allowing the bimorphs to cool and recover elastically to the start position. Control of wear is likewise crucial: in the ratchet devices illustrated in Fig. 10, or in their linear analogues (sometimes called 'scratch motors') loss of key geometry within the drive arrangement will lead to loss of function. The life of the wobble motor shown in Fig. 4 is similarly limited by wear, in this case of the central hub around which the rotor precesses in response to the changing sequence of electrostatic fields. Excessive wear at the centre can lead to contact at the circumference and so loss of function.
Fig. 11 Evidence of wear in a plain journal bearing in a silicon MEMS device [Tanner et al 1999].
81
Figure 11 shows a plain bearing in MEMS device before and after testing. The rotating journal carried a steady load and the resultant loss of silicon material is very clear - some of the finely divided debris can be seen around the edge of the bearing. The assembly of which this was part ran for 158,000 cycles; the shaft speed was in excess of over 100,000 rpm and this damage occurred in just over 90 seconds.
history. Napoleon's watch maker is reputed to have made the remark 'Give me a perfect oil and I will give you a perfect watch'. 1 m/s
1 mm
1 pm/s
length 1 [im
-:.-
3. TRIBOLOGICAL CHALLENGES IN MEMS In one sense thinking about friction, wear and lubrication at the small scale — the micro-scale — is not new to the tribological community. We have become accustomed to explanations of everyday macroscopic behaviour which depend on invoking microscopic or even molecular phenomena. For example, in order to explain the constancy of the coefficient of friction between macroscopic sliding components we reach for a model which requires a large population of small contact areas distributed within the nominal macroscopic geometric footprint of the contact and couple this with some rheological model of a surface film behaviour. Within the MEMS regime, the tribological aims remain reassuringly familiar - to control friction and minimise wear. Although MEMS devices are small they are micro- rather than nano. A 10 micron cube of silicon contains in excess of 1013 atoms so that there are unlikely to be dramatic changes in mechanical properties from tabulated or familiar values. Having said that, it is not wise to assume that there are no changes - a plot of characteristic length versus characteristic event time is instructive - Fig. 12. On log scales contours of constant sliding speed plot as straight lines with 45° slope. Engineering components will cover a length of perhaps from something less than a millimetre to metres and sliding speeds of less than Imm/s to perhaps 10 m/s. In MEMS devices the length scale of interest stretches from sub-millimetre to nanometres. Although the rotational speed in any MEMS motor is likely to be high, the sliding speed at the bearing surfaces is of the same order as in macroscopic devices. The diagram illustrates the fact that the particular challenge offered by MEMS design is that the technology is really just at the interface between two areas of professional expertise, viz. those of engineers and material scientists. This figure also illustrates how far from the area of MEMS operation are conditions, specifically sliding speed, within the Atomic Force Microscope. The devices illustrated in Figs. 1-11 represent a new order of smallness but the importance of friction and lubrication in fine machinery has a long
MICRO-
^ \
" " ' ' " • " "STRUCTURE
ATOMIC \
.'
STRUCTURE Jyimmic /
fn!
IUl
1,/s
/
/
•
A
MKMS.MMA->
"
/ 1 ms time
Fig. 12 Characteristic scales and contact times for various components and devices: typical operation of MEMS occurs close to the intersection of the areas associated with engineering and materials science. Tribological phenomena at this size range can have significant economic consequences. Moving coil instruments relied for their accuracy on low friction, low-wear pivots, often in the form of jewel bearings. During the 1930s the National Physical Laboratory carried out an extensive programme of experimental work into the design and operation of such bearings with particular reference to those found in electricity meters [Scott 1936]. One experimental run - to measure steadystate friction and wear performance - lasted more than 3.5 years. The results of this study were reported at the renowned General Discussion on Lubrication in 1937 [Scott 1937]. One of the first ideas that neophytes to tribology are made aware of is the notion that flat surfaces are not flat and that the tribological action happens at the tips of interacting asperities. This is (presumably) true at the MEMS scale and Greenwood and Williamson style models of surface roughness are much in evidence in the MEMS literature: there are, however, some additional difficulties in moving from the macro-world to the micro because the characteristic lengths that describe the nominal area of contact and the topography of the surfaces are scaled by different factors: this is compounded by the fact that the roughness values can vary from one poly layer to another and are different on the side walls of the etched trenches from values on the planar surfaces [Phinney 2004]. In addition, at this scale, surface energy terms may not be insignificant so that the Hertzian analysis must be extended to include their effect - for example, through either the JohnsonKendall-Roberts [1971] or Derjaguin-MullerToporov [1975] approaches as is appropriate.
82 3.1 Controlling friction in MEMS - Self Assembled Monolayers MEMS surfaces may come into contact unintentionally through unforeseen acceleration or electrostatic effects, or intentionally, when surface impact normally or shear past one another. If adhesive attraction exceeds the restoring forces, adhesion is permanent and the device has failed due to 'in-use stiction'. Approaches to reduce this can be classified under two headings, viz. physical and chemical modifications of the basic poly-silicon surface. In the physical approach the surfaces are roughened to reduce the effective contact area usually be selective etching, and effective reductions in surface energy as large as factors of 25 have been reported [Yee 1996]. In the chemical approach, the surface chemistry is altered to reduce the intrinsic surface energy and thus the propensity to stick. One technique is to treat the silicon surface with hydrofluoric acid which etches away the hydrophilic native surface oxide and terminates the Si with hydrogen. Since the Si-H bond is non-polar the resulting surface is hydrophobic and so has reduced in-use stiction. While this is attractive for devices encapsulated in inert atmospheres, it is known that the hydrogen terminated bonds are gradually re-oxidised in air so that the treatment is not permanent - although the effect may last several hundred hours which may be sufficient to provide commercial viability to the product. A more effective chemical surface modification involves the application of a mono-molecular film to the micromachine surface. Such Self-Assembled Monolayers, often deposited from solution, when correctly integrated into the manufacturing route, can provide a number of benefits. Firstly, they can effectively eliminate stiction due to meniscus effects during the release stage in wet-etch fabrication (by increasing the contact angle to more than 90° so rendering the surface hydrophobic rather than hydrophilic). Secondly, they can reduce the likelihood of the dry surfaces sticking (by a reduction in effective surface energy) and finally they may act as a boundary lubricant during use so reducing frictional losses and wear volumes. Table I Effect of SAMs on SiO2 surfaces Surface contact angle work of adhesion COF with water against Si degrees mJ/m2 SiO2 <30 20 -l(dry) OTS 110 0.012 0.07 Successful SAMs are robust organic molecules that are chemically adsorbed onto the Si or SiC>2 surface. Characteristically, they have a hydrophilic (polar) head that is bonded to the surface and a long,
hydrophobic tail that extends outwards. Octodecyltrichlorosilane (OTS) CigHsvSiClj is typical and has been intensively studied - its beneficial effects on SiO2 are indicated in Table I although it suffers from the potentially serious defect that the first step of the reaction sequence for binding the molecule to the surface involves the hydrolysis of the original Si-CI bonds thus liberating HC1 which could clearly be detrimental to any exposed metallic surfaces within the MEMS package. Perfluoropolyethers (PFPEs) are routinely used as relatively thick films as lubricants for magnetic storage media - both hard discs and flexible tapes. They have low surface tensions so reduce stiction, low vapour pressures and thus suffer very low evaporative losses and are oxidatively stable. The nature of the bond between the lubricant and the native silicon oxide surface depends on the end group of the PFPE - it may be non-polar (CF3) and therefore relatively weak as in Z-15- or polar (-OH) as in Z-DOL and can be thermally bonded by heating to 150°C for ca. 30 mins. The effect on friction - if not on long term wear - can be seen in Fig. 13. QKirictau of Diction 0.1 Si
0
tS
SO
75
100 125 Numlvr of cycles
(a) (b) Fig. 13 (a) Schematic of AFM tip sliding on Si(100) surface with native oxide (upper), non-polar Z-15 (mid) and polar Z-DOL (lower), (b) Coefficient of friction versus number of sliding cycles [after Liu and Bhushan 2003]. Much of the work into the frictional behaviour of SAMs and PFPEs (for example that illustrated in Fig. 13) has been carried out using an AFM as a 'scanning probe' device: the tip of the AFM probe is extremely fine so that even at very low loads the contact pressure is very much greater than any anticipated 'asperity pressure' in practice and (as has been noted in Fig. 11) the sliding speed is very low. This can be seen in a ' P V map as suggested by Dugger [2001, 2003] and illustrated in Fig. 14: here P is the likely maximum asperity pressure and V the imposed sliding speed. The Surface Force Apparatus [Homola 1989] can generate more
83 realistic sliding speeds but, as it relies on the contact between two relatively fragile mica surfaces, the load or pressure is limited. To some extent the first of these objections can be overcome, albeit with experimental difficulty, by sticking a small glass bead on the AFM tip to increase dramatically the effective radius of curvature; however, a number of commercial instruments are now becoming available which explore the more realistic region of the map designated 'SMM'. With MEMS, as in other situations where success depends on a degree of marginal lubrication, the boundary film must be sufficiently robust to survive for as many cycles of operation as is required - or else, some means of replenishment be provided - this was a significant concern in the design of the Tl Digital Micro-mirror Device. -re-
1000 800
I %
S o.
600
atomistic simulation
scanning probes surface forces apparatus
Fig. 15 Structure of (a) diamond, and (b) graphite, (c) The ternary phase diagram for amorphous carbon-hydrogen alloys.
~(iengine speed record
gear driving a load
400 200
"0
phases. Deposition methods have been developed to produce a-Cs with increasing sp3 bonding including the region of tetrahedral amorphous carbon designated ta-C. However, there is another important variable, viz the amount of hydrogen contained within the structure. This can vary from less than a few atomic percent to more than 50% and the extent of hydrogenation can be shown on a ternary phase diagram as in Fig. 15(c). Some coating methods, such as plasma enhanced chemical vapour deposition, are able to extend into the interior of the triangle producing hydrogenated amorphous carbon or a-C:H.
50
100 150 200 250
800
1000
Velocity, mm/s
Fig. 14 Experimental techniques, atomistic simulations and operating regimes of surface micromachines in pressure-velocity space. SMM refers to the Sandia sidewall tribometer referred to in the text [after Bugger 2001J. 3.2 Diamond like Carbon Films Diamond like Carbon (DLC) is a term used to describe a family of carbon based materials of whom many have attractive tribological properties - such as low friction coefficients (when slid dry against themselves or a variety of other materials) low wear rates, high values of elastic modulus, chemical inertness and good stability. Carbon can form a great variety of crystalline and disordered structures because it is able to exist in three hybridisations known as sp3, sp2 and sp1. In the sp3 configuration, as in natural diamond, a carbon atom's four valence electrons are assigned tetrahedrally as shown in Fig. 15(a): in the sp2, or graphitic, configuration, three of the four valence electrons lie in a plane, to which the final bond, now much weaker, is normal, Fig. 15(b). The so-called amorphous carbons, conventionally designated a-C, contain a mixture of sp2 and sp
Both types of film can, by appropriate choice of deposition conditions, be extremely uniform in thickness (with an intrinsic 'roughness' measured in tenths of nanometres) and thus, since they are only a few hundred nanometres in thickness, they retain the topography of the substrate on which they are laid down. Their mechanical properties can also be varied but in general ta-C has higher values of modulus and hardness than hydrogenated a-C:H as indicated in Table II. Table II Properties of DLC films compared to natural diamond [after Robertson 2002 and Grill 19971 ta-C a-C:H diamond H(at%) 0 30 sp fraction >0.8 <0.5 density (kg mf3) 3260 2350 3515 757 300 E (GPa) 1145 >20 <15 45 hardness (GPa) <1 residual stress (GPa) >6 These data also indicate one of the major problems with ta-C films, viz. the very high level of residual compressive stresses with which they are left. These can be annealed out at temperatures above ca. 400°C but this is unlikely to be feasible in a fabricated MEMS device. At the level of GPa there is a danger of the film delaminating from the substrate or distorting the components within the MEMS. A possible way forward is to create a nanostructured DLC layer by depositing a sequence of alternate harder ta-C and softer aC:H layers: in this
84 way it may be possible to build up a sufficient thickness for DLC to be used as a structural material rather than just a coating. Although the contact surfaces of such carbon films are chemically inert and so would not be expected to enter into major chemical activity with their environment they can still interact through physical adsorption and this mechanism can introduce significant fluctuations in friction and wear. For hydrogenated a-C:H the value of the COF when sliding against itself depends strongly on the relative humidity, see Fig. 16. Values below 0.05 are characteristic of dry air (or vacuum) but these rise by a factor three at more usual ambient values. The value of the COF also appears to be influenced by the ratio of H:C in the precursor gas used in the deposition. Tetrahedral ta-C and ta-C:H behave differently. They have higher COFs in vacuo or dry air but this seems to reduce - or at least not increase - in the presence of humidity.
Table III Super-lubncious DLC films [after Erdemir 2000] Carbon film COF grown in mm'N m 100%CH4 0.015 9.0x10" 25%H 2 +10%CH 4 0.01 7.36xlO'9 50%H 2 +10%CH 4 0.004 1.23xlO~8 90%H 2 +10%CH 4 0.004 2.8xlO"10 100%C2H2 0.27 7.5xl0~7 hydrogen free DLC 0.65 2.8x10"7 uncoated H13 0.8 4.6xlO~6 Microscopic examination of the DLC coated spheres and flats revealed very shallow wear racks and scars even after many hours of running: a longterm test of more 32 days was only terminated when the drive motor of the rig burnt out - the friction coefficient was consistently less than 0.005. The authors suggest that the remarkably low values of COF generated in this study are associated with the increasingly inert nature of the DLC surface with increasing availability of hydrogen during the deposition process. It is suggested that hydrogen passivated the dangling surface bonds that can otherwise lead to increases in friction and wear. 4. MEASURING FRICTION AND WEAR IN MEMS DEVICES
0.3 1 3 Relative humidity (%)
Fig. 16 Variation of friction coefficient with relative humidity for a-C:H and ta-C, after Robertson [2002], Enke [1981] and Voevodin [1996] When discussing the tribological properties of DLC, as with anything else, it is important to take account of the nature of the counterface. For a number of different counterface materials (including steel, silicon nitride and sapphire) long-term stability appears dependent on the formation of a carbon-rich transfer layer derived from the DLC but with a distinctive morphology of its own often consisting of fine graphitic nano-particles (<5 nm) within a distorted diamond-like structure. If both the extrinsic or environmental factors and the intrinsic properties of the film and the underlying substrates can be optimised then it is possible to generate extraordinarily low values of both COF and specific wear rates - see Table III. In these experiments the DLC films were deposited on extremely smooth (< 100 nm Ra) tool steel surfaces and tested in a ball on flat tribometer where the peak hertzian pressure was 1.04 GPa.
It is generally considered poor practice to run likeon-like surfaces in any tribological contact and so it is not perhaps surprising that silicon-on-silicon is no exception to this. The qualitative impression, for example the evidence of Fig. 11, is that wear rates in plain bearings in which two poly components run against one another are unacceptably high. If we are to allow for wear at the design stage we need some knowledge of the Archard wear constant - together, of course, with the assumption that this really is a 'constant'. Literature values of Kw for bare poly-onpoly at the macro-scale vary between 10" m m W 1 and 10~6 m m V m 1 [Gardos 1998]. The experimental determination of this number at the appropriate scale is not-straight-forward as we need to control load and sliding distance and measure wear volume loss within reasonable tolerance limits. It is difficult to extract data of this sort from experience with operational MEMS motors or MMAs because, in many cases, even if the sliding distance is known and wear volumes estimated from micrographs, the contact load is not measured or known with sufficient accuracy [Mehregany 1992, Williams 2001]. Several research groups are rising to the challenge to build a MEMS tribometer in which the necessary parameters are controlled and the desired variables measured. Figure 17 shows an example of
85 one such device, built by the Sandia Labs, which generates reciprocating motion, with stroke A, generated by flexing the poly-Silicon cantilever by means of electrostatic effects [deBoer 1998]. Actuation is performed under an interferemetric microscope so that deflections - and thus loads - can be assessed. An immediate problem with this arrangement is that A is small, only about 30 nm, so that it is not really feasible to assess wear quantitatively, but the nominal contact pressure can be varied from less than 10 kPa to about 25 MPa. To get useful work out of a functional MEMS or MMA will need nominal contact pressures of the order of a MPa. Observed static COFs between poly surfaces were between 2 and as much as 8: under sliding conditions this reduced very considerably to ca. 0.3 for silicon on silicon, and 0.16 for Si surfaces carrying a SAM.
suspended beam into contact with a fixed or anchored post under a known load, and the other to oscillate the beam against the post. The contact geometry is thus equivalent to a nominal line contact. Since the device itself is a MEMS the surface morphology and chemistry duplicate precisely those found in more complicated systems having contacting surfaces. The device operates in a PV range representative of other surface micromachined devices, see Fig. 14. A more recent device designed to characterize friction between micromachined surfaces is an 'nanotractor' device also developed at Sandia and shown in Fig. 19 [deBoer 2004]. It consists of a set of rails on the bottom of a moving structure that make contact with a planar track so that by repeating the sequence of electrostatically pulling one clamp down, deflecting the actuation plate, pulling the other clamp down and then releasing the first clamp and the actuation plate, the device can be walked out against the restoring force of the suspension springs and "load cell" beam. Static friction is determined by gradually releasing the clamping voltage until the device slips.
spring restoring force, lys
Fig. \1 (a) The Sandia 'Lab-on-a-chip' tribology tester in which the electrostatically driven flexure of the beam can be quantified by viewing by optical interference (b) [after deBoer et al 1998]. A similar device, which includes an additional link to extend the stroke is being used by a group at Dayton Research Institute [Patton 2000]. The device currently being used by the group at Berkeley is similar in some ways to the accelerometer illustrated in Fig.l. A planar shuttle, which carries a number of raised pimples on its lower surface, is actively reciprocated by a set of comb motors: simultaneously shuttle and substrate are pulled together by a second electrostatic arrangement so establishing a controlled load on the tips of the pimples. Nominal pressures are in the range 0.1 to 2.5 MPa and experiments can be carried out in air and in vacuum. 4.1 Sandia 'Side-wall' tribometer Sandia labs have built a further generation of devices to look specifically at sliding on the sidewalls of the chemically etched trenches on the basis that these are often the surfaces of interest in a functioning MEMS and are likely to have different topographies from those of the planar wafer surfaces. The device, the 'sidewall microtribometer' is shown in Fig. 18. It is driven by two electrostatic comb motors, one is used to pull a
^
ink
|
comb drive force. F c -Mo
Fig. 18 The Sandia 'side-wall' tribometer. Electrostatic actuators pull the suspended beam into contact with the anchored post while a second set oscillates the beam back and forth (Dugger 2003b). Dynamic friction is determined by releasing the clamping voltage and allowing the device to oscillate while measuring the displacement versus time. A damped oscillator model is used to determine the air damping and dynamic friction coefficients required to fit the displacement data. Parallel plate capacitors having relatively large area,
86 and the ability to tailor the size of the rails, enable investigation of a wide range of contact pressures.
Fig 19 The Sandia 'nanotractor device [deBoer 2004] 4.2 CUED micro-tribometer The Sandia devices illustrated above make use of the multilevel SUMMiT™ process route alluded to in §1. We have recently commissioned a test-rig in which the dimensions of the bearing under test approach the micro-scale and in which it is the only sliding contact downstream of any load-measuring element - but which is not itself a MEMS device. The test surfaces, which can be of silicon, form a thrust bearing between the appropriately profiled upper end of a vertical rotating shaft and a matching 'dimple' in a small circular disc which sits within a symmetrical 'top hat' supported by the shaft as illustrated in Fig. 20. The mass of the top-hat provides the load on the test bearing and its rotational speed is monitored by a non-contact probe.
the test thrust bearing. Initially this has been set up as a sphere of radius ca. 1 mm within a spherical dimple with a slightly greater radius. With different top-hats and shaft tips it is possible, by varying the nature of the bearing geometry, to generate effective maximum contact pressures from a few MPa to over 1 GPa. In principle, measuring the friction in this device is very straightforward. If the shaft is accelerated up to some speed and then the motor stopped the tophat decelerates at a constant rate given by the frictional torque divided by its polar moment of inertia - which is known. The geometry of the contact (ball in cup, ball in cone etc) provides an effective radius at which friction acts and thus the coefficient of friction between the two surfaces concerned can be calculated. To generate wear requires some slip between the driving and the driven surfaces of the bearing and this can be achieved by cycling the motor speed and counting the number of 'slip revolutions' that occur during each phase of acceleration and deceleration. However, a neater method of providing some retarding torque on the top-hat so that a steady-state is reached with the shaft rotating at one speed and the specimen moving relative to it at some fixed slip speed is to use the rim of the hat as the disc in an eddy-current disc brake. The specimens can be cut from a regular Si wafer and given the same sequence of surface treatments as a MEMS device so that the rig is appropriate for the evaluation of several of the suggested routes to tribological success such as laser texturing or the combination of hydrophilic and hydrophobic surfaces in a so-called half-wetted bearing [Spikes 2003]. 5. CONCLUSIONS
Fig. 20 CUED Micro-tribometer capable of generating contact conditions very similar to those within MEMS devices. The apparatus sits within a bell jar which can be evacuated so that the only torque acting to accelerate or decelerate the top-hat arises from friction within
• Micro-systems are rich with issues concerned with the properties, processing and mechanics of materials. Much of current MEMS technology is based on silicon and although this material has some attractive properties - particularly its fatigue resistance - it is less attractive tribologically. Alternative candidate materials must be compatible with MEMS fabrication routes. • Macroscopic machines are often limited by inertia effects: MEMS and MMAs are much more commonly limited by surface forces and the effects stiction/friction and wear. For MEMS, tribology is an important enabling technology. • Sliding speeds in MEMS are by macro standards comparatively modest - i.e. less than 1 ms~'. Sliding distances, in rotating bearings can be of order of hundreds of km if the device is likely to be in continuous use. • For 'reasonable' mechanical performance, nominal bearing pressures must be of the order of
87 MegaPascals and for acceptable life, specific wear rates need to be no more than ca. 1(T8 mm31ST1 m~'. • Within MEMS conventional liquid lubrication is impossible because of meniscus force effects. Self Assembled Monolayers applied to sliding surfaces may provide acceptable combinations of low friction and wear though robustness and replenishment may be issues of concern. • In dry running MEMS and MMAs, Diamond-LikeCarbon has some very attractive tribological features but some deposition routes can leave the film with unacceptably high values of residual stress. • Modelling the behaviour of small-scale contacts is challenging, not only because of the inclusion of surface energy effects but also because of the different factors by which component dimensions and surface topography may change as the scale of the device reduces. 6. ACKNOWLEDEGEMENTS Our thanks are due to Professor Victor Bright of the University of Colorado for Fig. 10 and to Dr Mike Dugger of the Sandia Labs for details of their micro- and nano-tribology facilities.
REFERENCES Bhushan,B (2003) J. Vac. Sci. Technol. B21 2262 Bustillo J M, Muller R S (1998) Trans IEEE 86 1552 Corwin A D, Walraven J A and Redmond J M (2004) J Mems 13 63 deBoer M P, Redmond J M and Michalske T A (1998) SPIE Proc. 3512 Santa Clara deBoer M P, Luck D L, Ashurst W R, Maboudian R, Corwin A D, Walraven J A and J M Redmond J M (2004) J Mems 13 63 Derjaguin B V, Muller V M and Toporov Y P (1975) J. Colloid Interface Sci. 53 314 Douglass TI White Paper: http://www.plusamerica.com/papers.html Dugger M T (2003) Quantification of friction in micro-system contacts in Nanotribology Critical Assessement and Research Needs ed. S M Hsu and ZC Ying, Kluwer, Boston Dugger M T (2003b) Tribology in Microsystems in TRIMS 2003: Microtribology Neuchatel, Switzerland. Elwenspoek M and Jansen H (1999) Silicon Micromachining CUP Encke K (1981) Thin Solids Films 80 227 Erdemir A, Eyrilmaz O L, Nilufer I B and Fenske G R (2000) Surface Coatings and Technol. 133134 448 Feynman R P(1992) reprinted in ASME J. Microelectromechanical Systems 1 60 Gabriel K J, Behi F, Mahadevan R and Mehegrany M (1990), Sensors and Actuators A21-A23 184
Gardos, M (1998) in Tribology issues and Opportunities in MEMS, ed B Bhushan, Kluwer, Dordrecht. Grill A (1997)Surface Coatings and Technol. 94-95 507 Homola A M, Israelachvili J N, Gee M L and McGuiggan P M (1989) Trans ASME J of Tribology 111 675 iMEMS® Integrated Micro-Electro Mechanical Systems, Analog Devices http://www.analog.com Johnson K L, Kendall K and Roberts A D (1971) Proc. Roy. Soc. A324 301 Linder C, Parette L, Gretillat M-A, Jaecklin V P and de Rooij N F (1999) J. Micromech. Microengng 2 122 Liu H and Bhushan B (2003) Ultramicroscopy 97 321 Mehregany M and Senturia S D (1992) IEEE Trans Electron Devices 39 1136 Miller S L and LaVigne G (1997) Proc SPIE Smart Electronics and MEMS 3224 24 Miller S L, Sniegowski J J, LaVigne G and McWhorter P J (1996) Proc SPIE Smart Electronics and MEMS 2722 197 Muhlstein C L, Stach E A and Ritchie R O (2003) Engng. Integrity 14 4 Patton, S T, Cowan W D, Eapen K C and Zabinski J S (2000) Trib Lett 9 199 Robertson J (2002) Materials Science and Engineering Reports, R37 129 Scott V (1936) National Physical Laboratory, Collected Researches 24 1 Scott V (1937) Proc IMechE General discussion on Lubricants and Lubrication, London 2145 Song X, Rajgolpal S, Melzak J M, Zorman CA and Mehegrany M (2001) Mater. Sci. Forum 389-393 755 Spikes H A (2003) Proc. Instn Mech. Engs 217J 1 Tanner D M, Miller W M, Peterson K A, Dugger M T, W P Eaton, Irwin L W, Senft D C, Smith N, Tangyunyong F P and Miller S L (1999) Microelectronics Reliability 39 401 Voevodin, A A, Phelps A W, Zabinski J S and Donley M S (1996) Diamond Rel Mater. 5 1264 Williams J A (2001)Wear 251 965 Yee Y, Chun K and Lee J D (1996) Actuators A52 145