The MEMS Handbook MEMS Applications (2nd Ed) - M. Gad el Hak Episode 1 Part 2 pot

0 1 2 3 4 5 6Field oxide Sac oxide BPSG TEOS PETEOS Nitride poly stud Mechanical poly Nitride Arsenic-doped epitaxial layer MM poly 0 n-type silicon substrate Micromechanical device area

(a) (b)

COLOR FIGURE 2.1 Examples of two high-volume accelerometer products Example (a) is the top view graph of the Analog Devices, Inc ADXL250 two-axis lateral monolithically-integrated accelerometer Example (b) is a perspective view of the Freescale Semiconductor, Inc wafer-scale packaged accelerometer and control chips stack- mounted on a lead frame prior to plastic injection molding (Photos courtesy of Analog Devices, Inc and Freescale Semiconductor, Inc.)

micro-COLOR FIGURE 2.4 Top view micrograph of a Z-axis accelerometer quadrant showing a folded spring and

sacri-ficial etch holes designed into the proof-mass structure (Photo courtesy Freescale Semiconductor, Inc.)

0 1 2 3 4 5 6

Field oxide

Sac oxide

BPSG TEOS

PETEOS

Nitride

poly stud

Mechanical poly

Nitride Arsenic-doped epitaxial layer MM poly 0

n-type silicon substrate Micromechanical device area

COLOR FIGURE 2.13 Cross-sectional diagram of the IMEMS process developed at Sandia National Laboratories demonstrating the transducer formed in a recessed moat and sealed prior to the commencement of the high density CMOS process (Photo courtesy Sandia National Laboratories.)

COLOR FIGURE 2.6 The frequency response x苶 versus normalized frequency ratio ω/ωn.

COLOR FIGURE 2.14 Top view micrograph of a Z-axis capacitive accelerometer in three polysilicon layers The design

allows for high inertial sensitivity with a low temperature sensitivity (Photo courtesy Freescale Semiconductor, Inc.)

0.00 0.05 0.10 0.15 0.20 0.25 0.30

r /w

3.0 2.8 2.6 2.4 2.2 2.0 1.8 1.6 1.4

K t

F F

W

W

r

W/w = 1.5 W/w = 2

COLOR FIGURE 4.10 Stress concentrations for a flat plate loaded axially with two different widths and fillet radius r.

The maximum stress is located around the fillets.

COLOR FIGURE 4.32 The gear teeth of the small gear are wedged underneath the teeth of the large diameter gear.

In this case, gear misalignment is about 2.5 mm in the vertical direction.

COLOR FIGURE 9.4 Schematic illustration of the capacitive charging: (a) and (b) demonstrate the electric field, and

F represents time averaged Maxwell force; (c) and (d) demonstrate the flow profile.

COLOR FIGURE 9.5 Schematic illustration of the Faradaic charging: (a) and (b) on the left, anions are driven to the

same electrode surface where cations are produced by a Faradaic anodic reaction during the half-cycle when the

electrode potential is positive; (c) and (d) the flow directions are opposite to those in Figure 9.4

(b)

(a)

COLOR FIGURE 9.7 Particle focusing lines along the stagnation points for capacitive charging The vertical force

toward the electrode is a weak DEP or gravitational force The circulation is opposite for Faradaic charging An actual image of the assembled particles is shown below.

COLOR FIGURE 9.8 The writing and erasure processes for Au electrodes at ω ⫽ 100 Hz The frames are taken at 0 s,

5 s, 10 s, and 15 s after the field is turned on The initial voltage is 1.0 Vrms and is increased to 2.2 Vrms at 7.0 s Particles

on the electrode in the first two frames (a) and (b) move in directions consistent with electro-osmotic flow due to capacitive charging and assemble into lines They are erased by Faradaic charging in the next two frames (c) and (d).

The arrows demonstrate the direction of particle motion The dashed lines are located at the theoretical L/兹2苶.

COLOR FIGURE 9.9 Bacteria trapping by AC electroosmotic flow.

COLOR FIGURE 10.11 Bubble volume variation versus time for three different heater designs under same heat flux

of 1.2 GW/m 2 , courtesy Yang, et al (2004).

COLOR FIGURE 11.7 Silicon wafer into which an array of micro heat pipes has been fabricated.

0 40 80 120 160 200 240 280

COLOR FIGURE 11.10 Temperature difference of micro heat pipe arrays with or without working fluid (Reprinted

with permission from Wang, Y., Ma, H.B., and Peterson, G.P (2001) “Investigation of the Temperature Distributions

on Radiator Fins with Micro Heat Pipes,” AIAA J Thermophysics and Heat Transfer 15(1), pp 42–49.)

0 500 1000 1500 2000 2500 3000 3500

COLOR FIGURE 11.11 Effective thermal conductivity of micro heat pipe arrays (Reprinted with permission from

Wang, Y., Ma, H.B., and Peterson, G.P (2001) “Investigation of the Temperature Distributions on Radiator Fins with

Micro Heat Pipes,” AIAA J Thermophysics and Heat Transfer 15(1), pp 42–49.)

tunneling tip, as well as low frequency noise sources, remain [Grade et al., 1996] Recently, Shashkin et al.(2004) proposed that Fowler–Nordheim tunneling-based inertial sensing could provide a more stable alter-native using parallel electrodes resulting in high sensitivity As tunneling-based technologies expand inapplication, researchers will find solutions to mitigate the current limitations of the methodology.

2.5 Rotational Inertial Sensor Parameters

Linear and rotational inertial sensors have much in common; for example, both exhibit a structure prising a specific mass as well as a flexible means by which this structure is anchored to the substrate, andboth types of sensors are often manufactured through the same or similar technologies Unlike a linear iner-tial sensor, however, the transducer of an angular rate sensor needs to be driven into oscillation in order togenerate a measurable signal (in most cases) This requirement comes from the coupling of vibratorymotion by the Coriolis Effect to produce a positional shift sufficient for sensing The requirement adds bothtransducer and circuit complexity to the system Upon a rotation of the transducer about its sense axis, a Coriolisforce is generated in the presence of a rotational velocity of the reference frame, which in turn drives the trans-ducer structure orthogonally as given in Equation 2.4 This means that a minimum of two orthonormalaxes of motion is required in order to suitably measure the small Coriolis force exerted on a resonating proof-mass during rotation Rollover sensors typically resonate in plane and measure normal to the surface Axes

com-of sensitivity for gyroscopic sensors are shown in Figure 2.8 The scalar governing equation com-of motion for a

gyroscopic device with a resonating mass in the Y-axis, rotated about the Z-axis is given by Equation 2.14,

⫹ 2ξωn ⫹ ω2

where Ωz is the rate of rotation and y is linear velocity of the structure due to the drive One may make

an analogy between rotational and linear sensors if the Coriolis term (2 Ωz dy/dt) is considered an

accel-eration According to a typical automotive spec where the full range of angular velocity is 100 deg/sec an

equivalent acceleration, a, is given by Equation 2.15,

In general, the driving frequency is near resonance and the vibration amplitude of the transducer ture is about 1 µm Assuming a natural frequency of 10 kHz, the resulting Coriolis acceleration of Equation2.15 has a value of 0.022 mg, demonstrating that this force-induced acceleration is very small

Resona

nt m ode

Cor

iolis

Acc

elerat n

z

x y

about x

Coriolis Acceleration

about y

M ass

For most applications, a single axis angular rotation measurement is required Such a single axis rate sensorcan be built by sensing induced displacement from an oscillating rotor or from a linearly oscillating structure.Although these two types of rate sensor designs appear to be very different, the operation principles are thesame In both cases, when the reference frame (or device substrate) experiences a rotation along the inputaxis, the oscillating mass (either translational or rotary), in a direction perpendicular to the input axis (referred

to as the drive axis), would induce a Coriolis force or torque in a direction perpendicular to both the inputaxis and the drive axis With the amplitude of the drive oscillation fixed and controlled, the amplitude of thesensing oscillation is proportional to the rate of rotation of the mounting foundation Feng and Gore (2004)show a mathematical model for the dynamic behavior of vibratory gyroscopes

Because the coupling of the Coriolis Effect is orthogonal to the vibratory motion in a micromachineddevice, two degrees of mechanical freedom are required One degree of freedom is utilized for the excitation ofthe vibratory motion, and the second degree of freedom orthogonal to the first is required for sensing Thisrequirement couples tightly into the technology choice for rotational inertial sensors, because the axis ofsensitivity defines which mechanical degrees of freedom are required to sense it For example, a very thick highaspect ratio technology — as is possible with direct wafer bonded structures — might not be the most suitablefor a device that is required to move out of the plane of the wafer However, as with their linear counterparts, mosttechnologies and sensing methodologies have been applied to vibratory sensors with new combinations

of methodologies always under consideration

Putty and Najafi (1994) provide a discussion of the varieties of rotational inertial sensors, including ing prismatic beams [Greiff et al., 1991], tuning fork designs [Voss et al., 1997; Hiller et al., 1998], coupledaccelerometers [Lutz et al., 1997; Kobayashi et al., 1999; Park et al., 1999], and vibrating shells [Putty andNajafi, 1994; McNie et al., 1999] As illustrated in Figure 2.9, He and Najafi (2002) demonstrate an all-silicon vibrating ring gyroscope with very good performance Multiple-axis systems have also beendemonstrated [Juneau et al., 1997; Fujita et al., 1997] In all cases, the vibrating structure is displaced orthog-onally to the direction of the vibrating motion This configuration can lead to system errors related to the

vibrat-transducer structure and the electronics The primary error related to the vibrat-transducer is called quadrature error and is discussed in the next sub-section.

As an alternative to single proof-mass designs, a concept involving two coupled oscillating masses hasemerged, with one mass for driving and one mass for sensing One of the first such designs is documented byHsu et al (1999), who used an outer ring as the drive mass and an inner disk as the sense mass The driving

mass is actuated by a set of rotary comb structures and oscillates about the Z-axis (or the vertical axis) The sensing disk is anchored to the substrate in such a way that the stiffness about the Z-axis is significantly greater

FIGURE 2.9 Perspective view scanning electron micrograph of a single-crystalline silicon vibratory ring gyroscope (Photo courtesy K Najafi, University of Michigan.)

than stiffness about the other axes The outer ring and the inner disk are connected by a set of flexiblebeams or linkages When there is no input of angular rotation, the oscillation of the drive mass about the

Z-axis has virtually no impact on the motion of the sense disk When the device experiences a rotation about either the X- or Y-axis, the Coriolis force-induced torque drives the inner disk into a rocking motion about the Y- or X-axis Electrode pads underneath the disk measure the variation of capacitance, which is pro-

portional to the input angular rate Another advantage of this two-mass design is that the dual proof-massstructure permits the ring and the disk to be excited independently so that each can be dynamically com-pensated for manufacturing non-uniformity

Several other vibrational devices have been demonstrated also involving two mutually perpendicularoscillating masses [Kobayashi et al., 1999; Park et al., 1999] In these designs, the drive mass is forced to oscil-

late along the Y-axis by comb actuators and the sense mass is forced to oscillate along the X-axis by a Coriolis force The magnitude of this Coriolis force is proportional to the input angular rate along the Z-axis The

angular rate of rotation is measured by detecting changes in capacitance with interdigitated comb tures attached to the sense mass The linkage between the drive and sense masses is designed in such a waythat the Coriolis force is transferred from the drive mass to the sense mass in an efficient way; yet the feed-back from the motion of the sense mass to the drive mass is kept to a minimum by frequency matching Acarand Shkel (2003) proposed a variation on this scheme using four masses in a decoupled mode between a two-degree of freedom drive oscillator and a two-degree of freedom sense oscillator to further reduce offsets andimprove the performance

Most of the earlier sensor designs involve a single proof-mass for both driving and sensing The proof-mass

is supported by a set of multiple slender beam linkages, usually made of the same semiconductor materials

as the proof-mass, to allow for movement in two mutually perpendicular axes A major drawback in a single proof-mass design is the cross-axis coupling between the drive axis and the sense axis, a phenome-non commonly referred to as quadrature error This coupling can be attributed to defects or small non-orthogonalities in the mechanical structure Because the sense displacement is a minute fraction of thetypical drive displacement, small structural defects can generate large quadrature errors in the system.Quadrature is compensated for by enhanced structural design, as will be demonstrated in the examplesbelow, as well as by the generation of quadrature canceling force feedback of position in the control elec-tronics [Geen, 1998] Fortunately, the quadrature error coupled from the drive vibration is 90 degrees out-of-phase with respect to Coriolis-induced vibration and can be phase-discriminated to a large degree in thecontrol circuitry at the expense of additional control circuit complexity However, the continued increasingcomplexity in the structural design in modern micromachined gyroscopes indicates that quadrature errorcancellation cannot be completely resolved by the control circuit

The sensitivity requirements for rotational inertial sensors far exceed those for most linear inertial systems,both in terms of the transducer design and the circuit complexity In vibrational systems, structural sensitiv-ity and absolute stability in the control electronics are required to accurately measure rotational rate Becausethe magnitude of the driven vibration is directly proportional to the magnitude of the Coriolis-induced out-put displacement, the structure and electronics are designed to maximize the coupling and stability of themagnitude Structurally, the driven oscillations can have large displacements, on the order of microns or eventens of microns in some cases The devices are also commonly operated in a near-vacuum environment

to minimize the impact of mechanical damping on the structure to maximize the resonant response, or the

Q, of the system Electronically, precise control of the driven vibration amplitude is paramount Phase

dis-criminating circuitry such as phase-locked-loop (PLL) control is used to drive the device displacement at ornear resonance to maximize the displacement while precise amplitude control is maintained Phase discrim-ination and synchronous phase demodulators are also required to sense the Coriolis force displacementand cancel quadrature effects [Geen, 1998; Kobayashi et al., 1999] As with accelerometer systems, the sensecircuitry can be operated in open loop or closed loop force feedback configurations to sense displacementwith the system tradeoffs discussed in a later section

2.6 Micromachining Technologies for Inertial Sensing

Micromachining technology, implemented to produce the transducer device, is coupled to the physicalprinciple used to sense the inertial displacement Comprehensive details regarding these technological devel-opments are described in Section II of this Handbook Bulk silicon micromachined technologies were firstimplemented for inertial sensors However, polysilicon-based surface micromachined technologies dominatethe current marketplace for micromachined inertial sensors The trend is to use higher aspect ratio “surface”micromachined technologies to produce inertial sensors

Surface micromachined capacitive inertial sensors were broadly demonstrated commercially as a result

of the collaboration between Analog Devices, Inc and the University of California-Berkeley in the duction of the Analog Devices, Inc iMEMS™ BiCMOS integrated surface micromachined accelerometerprocess technology, as shown in Figure 2.1(a) The technology embeds a 2 µm micromechanical polysiliconlayer into a BiCMOS process flow [Chau et al., 1996] Application of this process has more recently beenexpanded to gyroscopes, with the ADXRS150 utilizing a 4 µm structure [Lewis et al., 2003] Sandia NationalLaboratories [Smith et al., 1995], Motorola, Inc., Sensor Products Division [Ristic et al., 1992] (now FreescaleSemiconductor, Inc.), and Siemens [Hierold et al., 1996], among others, have all demonstrated industrialsurface micromachined inertial sensor technologies Limitations to surface micromachining are primarilyrelated to the technological challenges in producing low-stress, high aspect ratio structures that have demon-strated benefits for sensitivity, mechanical damping properties, and insensitivity to off-axis motion.Epitaxially deposited polysilicon eliminates the aspect ratio limitations of the standard LPCVD polysilicondeposition typically used in surface micromachining This technology, sometimes referred to as “epipoly”technology, also allows the monolithic integration of CMOS or BiCMOS circuitry with higher aspect ratio

intro-capacitive transducers, typically on 10–12 µm-thick epitaxial layers [Kirsten et al., 1995; Offenberg et al., 1995;

Geiger et al., 1999; Reichenbach et al., 2003; Baschirotto et al., 2003] Epitaxial deposition of silicon is competitive for micromachining to thicknesses of 50 µm These high aspect ratio transducer structuresare relatively insensitive to out-of-plane motion and provide suitable mechanical damping at reasonablepackaging pressures This material has desirable film properties with nearly immeasurable intrinsic stressand a high deposition rate [Gennissen and French, 1996]

cost-With a reasonably flexible interconnect scheme, epipoly technologies have demonstrated monolithic gration with CMOS and BiCMOS circuitry as well as device thicknesses ranging from 8 µm to over 50 µm.Challenges for this technology include that co-deposited epitaxial silicon and polycrystalline silicon havedifferent deposition rates that complicate fabrication High temperature polycrystalline films typical of epi-taxial deposition also suffer from severe surface roughness and very large semi-conical crystalline grains.Solutions to many of these issues have been documented by various sources [Kirsten et al., 1995;Gennissen and French, 1996; Bergstrom et al., 1999]

inte-Direct wafer bond (DWB) technology has long demonstrated the successful incorporation of thickcapacitively- or piezoresistively-sensed inertial sensor structures Recent advances in this technique havedemonstrated improved device interconnect through the use of a silicon-on-insulator (SOI) handle waferwith defined interconnect [Ishihara et al., 1999] Piezoresistive elements have also been incorporated on thesidewalls of very high aspect ratio DWB structures to provide transducers with both piezoresistive andcapacitive sensing mechanisms [Partridge et al., 1998] DWB transducer technologies provide great processand device flexibility Very high aspect ratio structures are possible, approaching bulk-wafer thicknesses ifnecessary, providing excellent out-of-plane insensitivity and mechanical damping properties Monolithicintegration with CMOS is also possible The technology requires significant process capability to suc-cessfully produce DWB structures at high yield

As SOI microelectronic device technologies gain popularity for high performance mainstream CMOSprocess technologies, the substrate material required for micromachining becomes cost competitive withalternative transducer technology approaches, making SOI more appealing for inertial sensing applications.SOI technology, as a descendant of DWB technology, provides technological flexibility with desirable deviceproperties, including the out-of-plane insensitivity and high damping associated with high aspect ratio

structures SOI technology provides the advantage of single-crystal silicon sensor structures with verywell behaved mechanical properties and extraordinary flexibility for device thickness, as with DWB tech-nologies Thicknesses can range from submicron to hundreds of microns for structural layers UnlikeDWB, SOI technologies often lack the flexibility of pre-bond processing of the handle and active wafers

to form microcavities or buried contact layers that are often implemented in DWB technologies Anothertechnology hurdle has been the choice of methodology to minimize parasitics to the handle wafer Even

so, SOI has demonstrated a significant increase in its popularity as a micromachining substrate[Delapierre, 1999; Lemkin, Juneau et al., 1999; Park et al., 1999; McNie et al., 1999; Noworolski and Judy,1999; Lehto, 1999; Usenko and Carr, 1999] Lemkin and Boser (1999) demonstrated the monolithic inte-gration of SOI inertial sensors with CMOS While technological hurdles still need to be overcome forbroad industrialization of SOI MEMS devices, the technology holds great promise for a broad techno-logical platform with few limitations Macdonald and Zhang (1993) at Cornell University developed aprocess known as SCREAM (Single Crystal Reactive Etching and Metallization), which produces SOI-likehigh aspect ratio single crystal silicon transducers using a two-stage dry etching technique on a bulk sil-icon substrate This technology had been limited by the difficulty in electrically isolating the transducerstructure from the surrounding substrate Sridhar et al (1999) demonstrate that this technology can nowproduce fully isolated high aspect ratio transducers in bulk silicon substrates With fully isolated struc-tures, this technology can produce 20:1 aspect ratio devices for thicknesses to 50 µm and can be mono-lithically integrated with circuitry Xie et al (2000) and Yan et al (2004) have also demonstrated a relatedtwo-stage release methodology on capacitive inertial devices formed in a CMOS integrated technology.This technology shows promise for full integration of high aspect ratio lateral inertial structures withCMOS Also, Haronian (1999) demonstrated an integrated FET readout for an inertial mass releasedusing the SCREAM process

The development of metal micromachined structures by electroforming has demonstrated 300 µm thicknickel structures with submicron gaps formed using LIGA (Lithographie, Galvanoformung, Abformung)techniques [Ehrfeld et al., 1987] LIGA-like process techniques using reasonably high-resolution thick UVphotoresist processes, have resulted in inertial sensor development in nickel, permalloy, and gold post-CMOSmicromachining [Putty and Najafi, 1994; Wycisk et al., 1999] Very thick structures are possible using thesetechniques with effective “buried” contacts to the underlying circuitry in the substrate As a single-layerprocess addition, the technology adds minimal cost to the overall sensing system High aspect ratio structuresare possible However, the material properties of electroformed materials are difficult to stabilize and can beprone to creep

Traditionally a bulk micromachined technology, the application of deep anisotropic etching of oriented silicon wafers has produced novel inertial sensors with very high aspect ratios Aspect ratios up

(110)-to 200:1 have been demonstrated using this technique, although the practical application of the ogy may limit the maximum aspect ratio to below 100:1 [Hölke and Henderson, 1999] This technologyoffers an elegant solution providing extremely high aspect ratios compared to anisotropic dry etchingtechniques The technology is somewhat limited in its application flexibility because the deep trenches arecrystallographically defined by the intersection of the (111) planes with the surface of the wafer and must

technol-be arranged in parallelograms in the etch mask Circular and truly orthogonal structures are not easilyconfigured for this technology

2.7 Micromachining Technology Manufacturing Issues

The manufacturability of a transducer structure should be considered as important as the performance

of the device In theory, a sensor structure may be designed as sensitive as desired, but if the structure not be manufactured in a robust manner, the effort is futile Issues such as release or in-use stiction, sta-bility of material properties, and the control of critical processes in the manufacture of inertial sensorsshould be investigated and understood The impact of high aspect ratio technologies creates new chal-lenges in controlling and maintaining processes

2.7.1 Stiction

Stiction is a term used in micromachining to describe two conditions: release and in-use stiction Releasestiction is the irreversible latching of some part of the moveable structure in the device caused during therelease etch and drying processes In-use stiction is the irreversible latching of the moveable structure dur-ing device operation All high aspect ratio technologies require a step to release the moveable structurefrom the supporting substrate or sidewalls at some point in the process flow This process is not always,but is most often a wet etch of a dielectric layer, using a solution containing hydrofluoric acid and water.Release stiction typically occurs during the drying step following a wet solution process as the surface ten-sion forces in the liquid draw the micromachined structure into intimate contact with adjacent surfaces Theclose contact and typically hydrated surfaces result in van der Waals attraction along the smooth parallelsurfaces, bonding the layers to each other [Mastrangelo and Hsu, 1993] Too large a proof-mass or too soft

a spring may dramatically increase the probability of stiction resulting in yield loss

There are many techniques employed to reduce or eliminate release stiction Supercritical CO2dryingprocesses avoid surface tension forces completely and often result in very good stiction yields However,this technique has been difficult to implement in industrial process conditions Surface modifications, oftenbased on fluorinated polymer coatings, have been used to reduce surface tension forces on the microma-chined structure during release and drying with some success As hydrophobic materials, these monolayercoatings require significant surface treatment and have not found broad industrial utilization yet Other tech-niques have been employed with some success, all at the cost of additional process complexity and struc-tural compromises The latest trend has been to utilize dry release processing, often related to a DRIE-lastprocess flow for a high aspect ratio device that is exposed through the substrate [Amini et al., 2004] However,the problem with stiction yield loss is increased with aspect ratio because the surface tension forces act over

a larger area High aspect ratio structures must be designed with care to minimize the complications fromrelease stiction

In-use stiction issues also increase with the aspect ratio of a device For capacitive accelerometers, theproof-mass closing in on an actuated electrode can cause electrostatic latching if the electrostatic force

becomes larger than the elastic spring’s restoring force This condition is called pull-in or electrostatic ing and is design dependent High aspect ratio designs result in more capacitive coupling force for a given

latch-device topography and can be more prone to latching Many latch-devices are designed with over-travel stops toreduce the risk from this compromising situation

2.7.2 Material Stability

While providing design performance and off axis stiffness, high aspect ratio devices remain sensitive tothe stability of material properties This is particularly important for polycrystalline silicon devices, since thedeposition process can result in variations in the average intrinsic stress of a film as well as generate stressgradients throughout the sensor layer However, all associated materials result in significant impacts on thedevice performance and repeatability Stability and uniformity of backside film stacks, plasma-assisteddeposited dielectric films, and even the proximity of metallizations in the front-end process can impact theuniform and controlled behavior of a device

As was mentioned previously, there are many advantages to using thicker structures for both linear and tional inertial sensors However, high aspect ratio silicon structures require low-stress structural layers aswell as deep etching capability The former issue is not of concern for bulk micromachined devices, but ifthe structure is to be formed from a deposited film, there are trade-offs among the various deposition meth-ods and conditions in order to obtain a uniform, smooth layer free of stress-induced curvature, particularly

rota-if a reasonably high deposition rate is desired Many of the common challenges associated with surfacemicromachining are exacerbated as the thickness of the structural layer is increased As previously stated,

epitaxial deposition of polysilicon structures results in very low-stress films with a high deposition rate,but additional measures are often required to reduce surface roughness Another solution involves thedeposition of polysilicon into trench-based forms [Chae et al., 2004] Both of these methods typically involvehigh-temperature processing, which may impose restrictions on fabrication sequencing, as will be dis-cussed in a later section.

Deep etching for high aspect ratio structures has taken on several different forms based on the desiredshape and uniformity of the resulting trenches In the case of (110) silicon technologies, a wet anisotropicetchant utilizes the etch rates along different crystallographic planes in the device material to control the pro-file of the trenches formed Control of such processing requires accurate alignment of the etch mask to thecrystallographic planes in order to successfully control the aspect ratio to a designed parameter [Hölkeand Henderson, 1999] The technique is also sensitive to impurities in the crystal In most high aspect ratiotechnologies, however, a deep dry reactive ion etch (DRIE) of the trenches forms the structure Deep trenchetching has been implemented using various techniques, but a process pioneered by Bosch [Laerme et al.,1999] in which the film is cycled between modes of reactive etching and sidewall passivation has demon-strated a clear predominance as an alternative Control of the etch properties and profile is the most signif-icant challenge for high aspect ratio technologies Many potential process conditions can degrade the etchprofile or complicate the uniformity of the process for across-wafer and wafer-to-wafer variations in theprocess These variations in profile and width strongly impact the design parameters such as spring constantand damping, etc

2.7.4 Inertial Sensor Packaging

Package interactions are just as critical as device technology choices and often contribute significant formance shifts from package to package [Li et al., 1998] Micromachined inertial sensors, while robust on

per-a micro scper-ale, per-are frper-agile per-at the per-assembly scper-ale per-and eper-asily dper-amper-aged per-and often require two levels of pper-ackper-aging:(1) wafer level packaging, which is usually hermetic to provide damping control and to protect the MEMSdevices from the subsequent assembly operations; and (2) conventional electronic packaging of die-bonding,wire-bonding, and molding to provide a housing for handling, mounting, and board level interconnec-tion The package must fulfill several basic functions: (1) to provide electrical connections and isolation, (2)

to dissipate heat through thermal conduction, and (3) to provide mechanical support and isolate stress Anindustrially relevant packaging process must be stable, robust, and easily automated, and must take testa-bility into account

Wafer level packaging techniques include silicon-to-glass anodic bonding [Dokmeci et al., 1997], compression bonding using glass frit [Audet et al., 1997] or eutectic [Wolffenbuttel and Wise, 1994; Cheng

thermo-et al., 2000], direct wafer bonding [Huff thermo-et al., 1991], and monolithic capping technologies [Burns thermo-et al., 1995]utilizing one or more wafers These techniques allow the transducer device to be sealed at the wafer level

to protect the movable components from damage during assembly The wafer level package also provides thesensor with a controlled ambient to preserve the damping characteristics of the proof-mass.Figure 2.1(b)shows a Freescale Semiconductor, Inc accelerometer die with a wafer level cap in silicon mounted on top

of the co-packaged CMOS control IC

The unique challenge of sensor packaging is that in addition to providing a mounting foundation to a

PC board, one must control stresses that are induced by mismatch in the thermal expansion coefficients ofthe materials used to fabricate the package and the external thermal loading of the package These stressesmust be kept at a level low enough to avoid impact to the sensor or control circuitry performance Anexample of this challenge is illustrated in Figure 2.10, adapted from Li et al (1998), where external package-induced stresses on an accelerometer die produced a 0.15 µm curvature out-of-plane for the die from cen-ter to edge, resulting in a device offset that would not be present for the die in wafer form For capacitivedevices capable of resolving displacements at the nanometer or sub-nanometer scale, excessive curvature due

to stress on the die at a late point in the assembly can be catastrophic In general, different MEMS deviceshave different stress tolerance levels Therefore, each MEMS package must be uniquely designed and eval-uated to meet special requirements [Dickerson and Ward, 1997; Tang et al., 1997]