`
`
`
`Sensors and Acw:uors A 56 ( 1996) l67-177
`
`
`
`
`
`Polysilicon integrated microsystems: technologies and applications
`
`
`
`Roger T. Howe•. Bernhard E. Boser•, Albert P. Pisano
`
`b
`
`
`
`
`
`
`
`
`
`' Uni�enity nf Ci,liforn;a nt Berkel"!y, BerU'ley Sen:mr & Actuator Center. lH{'(lrtmenl 11/ Elutricat £ngmunn,: ar.d Computer Sdences. 497 Cory Hall.
`
`
`Btrkti,y. CA 947ZIJ-177<. USA
`
`
`
`
`
`
`
`Berk,/,y, CA 947Z0-!77', USA
`
`
`
`h Ut1iwnity,ifC,d1fr,11,1a ar Bnk.e/ey, Berkeley Sensor d: Actuawr Ceruu. De(1drtmn11ofMe,·h(Jnicul Enginuring, 5101-B Erchew!rry.
`
`Abstract
`
`Co-fabrication of polysilicon microstructures with CMOS electronics enables monolithic inertial sensors to be fabricated. Correlations arc
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`well established between 1he deposition, doping, and annealing conditions of LPCVD polysilicon and its mechanical prope1•ies, such as
`
`
`
`
`
`
`
`
`
`
`
`
`residual strain, strain gradient, and Young's modulus. Surface passivation, for alleviating stiction have been demonstrated recently that greatly
`
`
`
`
`
`
`
`
`
`
`reduce the tendency for poly silicon microstructures to adhere to adjacent surfaces when dried after release by wet etching and rinsing, or when
`
`
`
`
`
`
`
`
`brought into contact due to mechanic�! shock during use. Sigma�ella control stralegies are attractive for linearizing closed-loop sensors and
`
`
`
`
`
`
`
`
`are well suited to implementation in CMOS. Basic design principles for sense elements a11d electromechanical actuation in a singtc structural
`
`
`
`
`
`
`
`
`
`layer of polysilicon have emerged rapidly in the past several years. Monolithk polysilicon integrated sensors for the X, Y. and Z components
`
`
`
`
`
`
`
`
`or linear acceleration, angular rat-e, and angular acceleration have been demonstrated using the BiMEMS process of Analog Dc\rlces, Inc.
`
`
`
`
`
`
`
`
`
`
`Mechanical suspensions, electrostatic actuatm, and capacitive pickoffs, ar1 interface-and control-circuit building blocks arc all ponable to
`
`
`
`
`
`
`
`
`alternative inte�rated technologies that share 1he basic ch8J'3.cteristic of thin, laminar suspended microstructures.
`
`
`
`
`
`K�ywo.rds · lnlcgralec1 microsyslems � Polysilicon
`
`I.Introduction
`
`This paper begins by reviewing polysilicon micromachin
`
`
`
`
`
`ing technology and the approaches to co-fabrication of poly
`By selectively etching sacrificial layers from a multilayer
`
`
`
`
`
`
`silicon microstructures with CMOS. The control of the
`
`
`
`sandwich of patterned thin films, micromechanical structures
`
`
`
`mechanical properties of polysilicon has been studied exlen
`
`
`
`can be fabricated using integrated-circuit (IC) equipment and
`
`
`
`
`sively and is relatively well understood. Although the initial
`
`
`processes. This surface-micromachining process was first
`
`
`
`
`commercial inlegrated technology, Analog Devices' Bi
`
`
`used to make air bridges for low-capacitance interconnections
`
`
`MEMS process [ S], inlerleaves micromachining and elec
`
`
`in high-frequency !Cs. Micromechanical beams were co-fab
`
`
`
`
`
`tronics processmg steps, progress has been made recently in
`
`
`
`ricated with simple MOSFETs in the resonant-gate tr,msistor
`
`
`
`modularizing the fabrication sequence. In the second part of
`
`
`
`in the mid-I 960s [ I ] . In the early 1980s, surface microma
`
`
`the paper, several integrated sensors are described that have
`
`
`
`chining was applied to make polysilicon microstructures,
`
`
`
`recently been demonstrated in BiMEMS, including linear
`using an SiO2 sacrificial
`
`layer [ 2,3 I. Due primarily to the
`
`
`acceleromelers, angular acceleromelers, vibratory rate gyro
`
`
`
`
`difficulty of off-chip detection of the motional current from
`
`
`
`scopes, and resonant accelerometers. These initial results
`
`
`
`
`
`a vibrating microbridge, a simple process to integral� NMOS
`
`
`
`
`cs1ablish lhe feasibility of monolithic multi-sensing !Cs thal
`
`
`
`electronics with poly�ilicon microstructures was der.ion
`
`
`
`will have many applications in, for example, vehicle control
`
`
`
`strated in 1984 [ 4]. Polysilieon micromechanics was increas
`
`
`
`
`and head-mounced display systems. Furthennore. the micro
`
`
`
`
`ingly recognized as a promising technology through the
`
`
`electromechanical building blocks for the sense ekments,
`
`
`
`1980s. A major contributing factor to the growth of polysil
`
`
`
`such as suspensions and position-sensing capacitors. 1ogether
`
`icon micromachining was the mature
`
`infrastructure for depos
`
`
`
`
`with the interface and control circuits, can be captured in
`
`
`
`iting, pallerning, and etching thia films that had been
`
`
`
`CAD models and reused in new designs, in the same way as
`
`
`
`established by the IC industry. Polysilicon integrated micro
`
`
`
`is done in integrated circuits. fl.I present, there is much room
`
`
`
`systems are now being successfully commercialized, starling
`
`
`for improving the sophisticatirn and ease of use of CAD for
`
`
`
`with a 5Gg accelerometer for air-bag deployment introduced
`
`integrated microsystems
`in 1993 [5].
`
`/96/S 15.00 Ci 1996 Elsevier ScienCf! S.A. All righ1s
`
`resented
`0924-4:!47
`
`Pl/ S0924-424 7 ( 96) 01291 •5
`
`APPLE 1117
`Apple v. Logan Tree
`IPR2022-00037
`
`1
`
`
`
`168
`
`R.T. Howe el al. / Sensors and Actuators A 56 (1996) 167-177
`2. Polysilicon micromachining technology
`The fabrication sequence ofa polysilicon lateral resonator
`[6,7] is useful to oudine the basic process. After describing
`the basic process mod,'~s, we shall summarize recent
`research in mechanical pr,.~erty control and in the elimination
`of stiction during the final drying step.
`Fig. 1 shows cross secti.~ns through the process sequence,
`starting with the formation of a substrate ground plane using
`an n + diffusion in Fig. l(a). The wafer is passivated with a
`layer of 1500 .~ thick LPCVD nitride deposited on top of a
`layer of 0.5 ~m thick thermal SiO2. Contact windows to the
`substrate ground plane are then opened. Deposition, defini-
`tion, and patterning of an in situ phosphorus-doped polysili-
`con interconnection layer follows in Fig.l (b). This layer
`serves as a second electrode plane and as the interconnection
`to the n + diffusion and the microstructure.
`A 2 /.tin thick LPCVD sacrificial phosphosilicate glass
`(PSG) layer is deposited and patterned in two separate mask-
`ing steps. The first is a timed etch to create dimples, as shown
`in Fig. l(c). The second masking step etches through the
`PSG layer in windows to become the anchors of the polysil-
`
`icon structure, as shown in Fig. 1 (d). The 2/~m thick poly-
`silicon structural layer is then deposited by LPCVD
`(undoped) at 610 °C in Fig. l(e). The structural layer is
`doped by depositing another layer of 0.3 /zm thick PSG
`(Fig. 1 (f)) and then annealing at 1050 °C in N2 for one hour.
`This step dopes d,e polysilicon symmetrically by diffusion
`from the top and the bottom layers of PSG, in order to achieve
`a uniform grain texture and avoid gradients in the residual
`stress. The top PSG layer is then stripped and the structural
`polysilicon is patterned by reactive-ion etching to yield the
`cross section in Fig. I (g). Finally, the wafer is immersed in
`]O:! diluted HF to etch the sacrificial PSG layer. The wafer
`is rinsed extensively in deionized water to form a native oxide
`passivation on the polysilicon. A brief H202
`dip is sufficient
`to form the hydrophilic native oxide layer. After drying under
`an IR lamp, the final cross section is as shown in Fig. 1 (h).
`2. !. Polysilicon mechanical properties
`Polysilicon as a mechanical material has been the subject
`of extensive study since the ! 980s [ 3,8 ]. Residual stress and
`its gradient through the thickness of the film are critical con-
`
`Undoped poly~illcon
`(2 pm)
`
`Silicon nttddo
`Cor~act
`(0.15 pro)
`window
`n+ diffusion
`Thermal oxide
`ground Diane
`(0.5 IJm)
`
`(a) '
`
`Doped polysifcon
`interconnect {0.3 izm)
`
`(b) i . . . . . . . .
`
`Dimple
`(1 tim)
`
`Sacrificial PSG
`(2 IJm)
`
`/
`
`Top PSG
`(0.3 pro}
`
`:--
`
`. .
`
`.
`
`.
`
`.
`
`.
`
`
`
`.
`
`--
`
`$ .
`
`
`
`.
`
`•
`
`..
`
`7-1
`
`/--~L~=~ -~--
`~--- ~-~
`:
`\:_ .~____/_.
`.
`.
`.
`.
`
`
`
`;
`
`I
`
`~ :] '
`
`PIMma olch
`F-"J-:~
`
`:'-7
`
`(c)!
`
`_
`
`~
`
`..
`
`I
`
`.
`
`.
`
`.
`
`.
`
`.
`
`
`
`..
`
`(g)
`
`Anchor
`
`Suspended
`plate
`
`Bump
`
`Comb
`fln~er Anchor
`
`[ ......
`
`~
`
`~: ~:~i~".
`
`(h)
`(d)
`Fig. I. Polysilicon surface-micromachining process seq~nce [7] (cou,esy of Dr W.C. "l'ang, Jet Propulsion Laboratory. Pasadena, CA. USA).
`
`2
`
`
`
`Release and drying of the microstructure is a critical step in
`the process, which may result in stiction of the structure to
`the substrate after rinsing and drying. Several approaches to
`stiction-free drying have been demonstrated, including
`freeze-drying [ 171, supercritical CO2 drying [ 18], and the
`addition of polymer spacers to allow the final release to be
`done in an oxygen plasma [ 19].
`Stiction due to contact with adjacent surfaces after release
`remains a fundamental reliability question for surface-micro-
`machined structures. In order to minimize the work of adhe-
`sion, it is desirable to form a "hydrophobic surface on the
`polysilicon. Hydrogen-terminated silicon is extremely hydro-
`phobic [20], as are self-assembled monolayers (SAMs)
`[21 ]. Recent results with SAM coatings have demonstrated
`that 2.i5/~m thick polysilicon cantilevers) which are spaced
`2 ~tm from the substrate and up to i ram in length, can be
`released with high yield when the wafer is pulled directly
`from the final water rinse, as shown in the scanning electron
`nlicrograph (SEM) in Fig. 2. When cantilevers arc electro-
`statically deflected into contact with an underlying poiysili-
`con electrode at the same potential, SAM-coated beams are
`found to have a 50% sticking probability at a length of 950
`~m, in comparison with 110/,tin for hydrophilic polysilicon.
`When these results are translated into works of adhesion, the
`SAM coatings have a work of adhesion of 3 /zJ m-2 in
`comparison with 20 mJ m -2 for polysilicon passivated with
`a hydrophilic native oxide [21 ].
`The design of the 'back-end' sequence of microstrueture
`release, anti-stiction coating process, drying, dicing and
`assembly, and, finally, hermetic packaging is highly con-
`strained. For example, an oxygen plasma release [ 19] will
`remove organic coatings such as SAM films, with the impli-
`cation that they must be deposited after microstructure
`release. Anti-stiction coatings or processes must be able to
`survive the thermal cycle required by the hermetic packaging
`process. SAM coatings are stable to at least 400 °C in a
`
`R.T. Howe et al. / Sensors and Actuators A 56 (1996) 167-177
`straints on microstructure design. If the average stress is com-
`pressive, then microbridges will buckle if longer than a
`critical length. Stress gradients generate an internal bending
`moment that causes cantilever beams to warp out-of-plane
`upon release. The Young's modulus and yield stress vary
`somewhat with processing, but the range of variation is small,
`compared with the residual stress.
`The basic correlations between residual stress and stress
`gradient and the texture of undopcd and phosphorus-doped
`polysilicon are now reasonably well understood [9]. From
`these fundament~ studies, several types of poly,.ilicon have
`been developed. T,1¢ fine-grained, undoped polysilicoa
`developed at the University of Wisconsin is deposited in a
`microcrystalline state at 575 °C. A low-temperature anneal
`results in little grain growth, but the residual ~train becomes
`tensile and is sta~,le, for further annealing cycles. Conducting
`regions are formed by ion implantation into this semi-insu-
`lating fine-grained polysilicon [ 10]. At Berkeley, the focus
`has been on in situ phosphorus-doped polysilicon with rapid
`thermal stress annealing, in order t~ reduce the impact on
`pre-processed CMOS [ I 1 ]. By reducing the phosphine flow
`and using a deposition temperature of 585-590 °C, a higher
`deposition rate is achieved without much in:rease in the sheet
`resistance. Low sheet resistance and un~. forra doping 0u'ough
`the polysilicon are desirable to minimize depletion effects on
`the linearities of sense and force-feedback capacitors [ 12].
`By using rapid thermal anneal~,,g at 950 °C for 7 rain, a tensile
`residual stress of less than 25 MPa with negligible stress
`gradient through the film thickness can be achieved [ 11 ].
`The typical widths and thicknesses of LPCVD polysilicon
`microstructures are both in the range of 1-2 btm. Due to
`variations in the deposition, lithography, and etching proc-
`esses, the run-to-run variations in the width and thickness are
`on the order of 10%, which results in substantial variations
`in the spring rates, masses, and the resulting resonance fre-
`quencies [ 13], even if the stress and other properties such as
`Young's modulus were perfectly controlled. In order to
`achieve a precise resonance frequency, electrical or mechan-
`ical post-fabrication trimming is necessary.
`It is worth emphasizing that polysilicon is a low-loss,
`extremely stable mechanical material. Quality factors of
`50 000-100 000 are typical of polysilicon microresonators
`[7,14]. Electrostatically driven polysilicon resonant struc-
`tures encapsulated in thin-film vacuum chambers have a
`short-term stability of better than 0.02 Hz, for a resonance
`frequency of 625 kHz [ 14]. Operation for over three yeats
`with less than 0.4 ppm long-term frequency variation dem-
`onstrates conclusively the suitability of polysilicon for pre-
`cision sensing applications [ 15].
`2.2. ~i,,ostructure release and surface pas~ivation
`The final step in surface micromachining involves wet
`etching of the sacrificial oxide in hydrofluoric acid. The basic
`understanding of this process has improved considerably and
`models have been developed to predict etching times [ 16].
`
`Fig. 2. Tips of SAM-coated poi~silicon cantilcve~ (2.15 ttm thick, off~
`2.0 ~
`above the substrate: longest is 1000 pan long) that have been dried
`by pulling the wafer directly from the water rinse. Note the near-zero gradient
`in the tgsidual ~
`fog this in situ phosphorus-doped polysilicon film ( 211.
`
`169
`
`_ ~'i ~
`
`3
`
`
`
`170
`
`R.T. Howe et al. / Sensor.~ and Actuators A 56 (I 996) 167-/77
`
`nitrogen ambient, but only to around 150 °C in air [21].
`Thin-film r::!croshell packaging processes are attractive for
`simplifying the back-end [ i0,22]; however, the high depo-
`sition temperatures of LPCVD films eliminate most options
`for reducing the work of adhesion of the encapsulated micros-
`tructures. Recently, a process has been developed for wafer-
`to-wafer transfer of moided-polysilicon caps for vacuum
`.... ckaging. Hermetic sealing is achieved using a Au-Si eutec-
`~c bond at a temperature of 370 °C, which is compatible with
`SAM coatings [ 23 ].
`
`3. Polysilicon integrated microsystems technology
`
`Technologies to co-fabricate polysilicon surface micro-
`structures and CMOS electronics are necessarily complex. In
`addition to the deposition and patterning steps required by
`the micromechanicai and electronic processes, there are steps
`inserted to protect one region from damage during processing
`of the other. As well as considering the thermal cycles and
`materials compatibility, the typical overall step height of a
`polysilicon process is 4-5 pm for a single structural layer and
`6-7 pm for a double-structural-polysilicon process [ 24 ].
`An interleaving of microstructure and electronics steps is
`perhaps the most straightforward approach [4,5]. Most
`micromachit~ing steps are executed after the completion of
`the CMOS structv"es, but prior to metallization with alumi-
`num. The BiMEMS process by Analog Devices integrates
`both bipolar and MOS transistors along with a single-struc-
`tural-layer polysilicon microstructure process [ 5 ]. An gEM
`of a BiMEMS accelerometer, the ADXL-05 low-g device, is
`shown in Fig. 3 [ 25 ].
`In order to avoid threshold voltage shifts, the polysilicon
`microstructure anneal must be limit::d to under about 950 °C.
`The in situ phosphorus-doped polysilicon process discussed
`earlier [ I 1 ] is compatible with this temperature ceiling. The
`
`........
`
`, .
`
`.
`
`.
`
`.
`
`
`
`~
`
`.
`
`!:
`
`Fig. 3. SElvi of Ih~ Analog Devic~ Ai)Xi.,-05 acceieromelcr 12"/1. The
`structural polysilicon layer is 2/~m thick and suspended 1.6 ~m over the
`substrate. (Courtesy of Dr K.H.-L. Chau, Analog Devices. Inc.. Wilmington,
`MA, U~A).
`
`contact and metallization steps require modifications due to
`the rather severe topography created by the thin-film stacks
`of PSG and structural polysilicon. Photoresist is used to pro-
`tect the circuit area during the hydrofluoric acid etching of
`the sacrificial PSG layer 14].
`There are several benefits if a modular fabrication process
`can be designed, in which the micromachining and electronic
`steps are separated. Th,,~ MICS process, developed at UC
`Berkeley in the early 1990s, fabricates a modified p-well
`CMOS prior to polysilicon micromachining [ 24,26]. In order
`to raise the temperature ceiling of the CMOS, tungsten is used
`instead of aluminum for the metailization. TiSiJTiN diffu-
`sion barriers prevent WSi 2 formation during the subsequent
`Si3N4 deposition at 830 °C and rapid thermal stress-annealing
`step at 950 °C. Although MICS was used to demonstrate
`several integrated microsystems, subsequent research at San-
`dia National Laboratories showed that it ~s not sufficiently
`robust for manufacturing [27]. High and variable contact
`resistance to the p+ diffusions after the polysilicon micros-
`tructure thermal cycles, as well as delamination of tungsten
`metal lines, remain problems.
`Recent work at Sandia National Laboratories has estab-
`lished that a micromechanics-first integrated technology is
`feasible [ 28]. The key innovation is to bury the p..91ysilicon
`microstructures in a well etched into the substr~te, after which
`an overfilled oxide deposition is chemical-mechanical pol-
`ished (CMP) to planarize the wafer. The sche,-natic cross
`section of the single-structural-layer process is shown in
`Fig. 4. Studs of mechanical polysilicon are used to intercon-
`nect the CMOS alumir um metailization with the first poly-
`silicon layer in the well (MM Poly 0 in Fig. 4). In th~s
`process, the mechanical polysilicon can be annealed at high
`temperature in order to stabilize its properties against the
`thermal cycles of the CMOS fabrication steps.
`Advanced CMOS processes (e.g., with multiple levels of
`metailization) can be used without changing the microme-
`chanical process flow. Alternatively, a deeper well would
`make feasible multiple structural polysilicon layers. If deep
`reactive ion etching (RIE) patterns are refilled with sacrificial
`and structural layers, then three-dimensional st~'uctures can
`also be integrated. Using this approach, a 30 pm high molded
`polysilicon proof mass was embed0ed in the wafer, prior to
`conventional surface micromachining [29]. Modularity is
`therefore very attractive, since it enables separate develop-
`ment paths for the CMOS and polysilicon MEMS, which will
`be necessary for high-performance integrated microsystems.
`Fi,mlly, a different approach to polysilicon integrated
`mierosystems is the ¢pi-poly technology [ 30]. The mechan-
`ical material is polysilicon formed in an epitaxial silicon
`reactor over a patterned layer of SiO2. Since the deposition
`temperature is high, the epi-poly film deposits rapidly to a
`thickness of 10 pro. The CMOS electronics are fabricated in
`the silicon epitaxial layer surrounding the island of epi-poly.
`This process has the advantage of p~oviding thick structures
`with narrow gaps. Since it provides a single structural layer
`with high-aspect-ratio features, it can be classified with the
`
`4
`
`
`
`R.T Howe et aL / Sensors and Actuators A 56 (1996) 167-177
`
`171
`
`CMOS Device Area
`
`Micromechanicai Device Area
`
`. . . . . . . .
`
`a , , . . _
`
`~
`
`m , . . . _
`
`PC~t2
`
`u
`
`n-typa atlk:on m
`Fig. 4 Cross-sectional schematic of the embedded MEMS integrated technology [281 (courtesy of Dr J.H. Smith. Sandia National Lalxntot'ies. AIImq~rque.
`NM, USA).
`deep-RIE single-crystal silicon integrated processes to he
`discussed later.
`
`4. Polysilicon integrated inertial sensors
`In a merged surface-micromachined polysilicon micro-
`structure+CMOS process, a variety of MEMS building
`blocks are available from which to design closed-loop inertial
`sensors. Interdigitated comb structures ate usefu! for lateral
`and vertical electrostatic actuation, as well as for position and
`velocity sensing. The sense capacitances are typically on the
`order of 100 IF, which requires that great care be taken in
`designing the front-end amplifier.
`Fig. 5 is a die photograph of the second BiMEMS multi-
`project run, which was completed in June 1995. In this Sec-
`tion, we describe a variety of inertial sensors designed in the
`Analog Devices BiMEMS technology by our group at
`Berkeley.
`4.1. Linear accelerometers
`The initial application of the MICS and BiMEMS tech-
`nologies was to linear accelerometers [ 26,3 ! ]. With only a
`single structural layer, it is challenging to design a closed-
`loop accelerometer that is sensitive to the Z-axis component
`of acceleration, which is defined to be perpendicular to the
`surface of the substrate. The interdigitated comb is employed
`to provide a d.c. levitation force, which is balanced by the
`downward force supplied by a parallel-plate capacitor formed
`by the suspended proof mass and an underlying electrode.
`This sensor architecture was demonstrated in MICS [26] and
`fully implemented in BiMEMS [ 32].
`The asymmetrical nature of the Z-axis accelerometer
`makes it essential to use a digital-feedback scheme, since
`
`matched differential capacitors are not required in order to
`linearize the quadratic electrostatic fe,,~back force. Using a
`sigma-delta controller, a one-bit decision is made every clock
`cycle on the proof-mass position; the output signal is a serial
`bit stream at the clock frequency [33]. A simple level-shift
`circuit translates the bit stream into electrostatic force pulses
`that balance the suspended proof mass against deflections
`caused by input accelerations. As the clock frequency is
`increased, various artifacts of the quantization process are
`reduced. Digital signal processing is necessary to decimate
`the oversampled output signal and recover a digital represen-
`tation of the input acceleration signal.
`Lateral components of acceleration, which arc in the plane
`of the substrate, can be detected using a suspended structure
`incorporating differential interdigitated sense and feedoack
`combs such as the ADXL-05 shown in Fig. 3. The electro-
`static force can be linearized using analog techniques in this
`case, since the differential capacitor is well matched How-
`ever, the sigma-delta feedback loop is still attractive, since it
`does not require precision analog signal proce~ing and is
`suitable for implementation in digital CMOS. In addition, the
`sampled nature of the position measurement makes it ame-
`nable to canceling drift in the electronic circuit by subtracting
`the initial offset and errors due to charge injection.
`Fig. 6 is a die photograph of a fully differential lateral
`accelemmeler using sigma-delta control of the proof-mass
`position [3,:,35]. The capacitive sensing on this accelero-
`meter is done by applying a pulse waveform to the sense
`element and detecting the charge imbalance in a common-
`centroid layout of differential capacitors, which is opposite
`to the method used in the ADXL-50 [31] and ADXL-05
`[25]. As a result, the sense amplifier is fully differential,
`which helps to eliminate offsets.
`For a sarnpiing frequency of 500 kHz and a resonance
`frequency of 8 kHz, the sensor achieves a noise floor of 500
`
`5
`
`
`
`i 72
`
`R T. Howe et al,/Sensors and Actuators A 56 (1996) 167-177
`
`Pig. 5. Second BiMI-MS multi-project chips, including designs from UC l~:tk~;ley, MIT, and Stanford (courtesy of Dr R.$. Payne, Analog Devices, Inc.,
`Wilmington, MA, U3A).
`
`F8 Hz- !/2. Electro.ic noise from the digital-to-analog con-
`vener used to generate the rebalance voltage levels is the
`dominant contributor. The fundamental input-referred elec-
`Ironic noise floor is that due to the sense amplifier, which can
`be on the order of 1 /.tg Hz-~/2 in a I /~m channel length
`CMOS technology [33]. Operation in a vacuum would be
`necessary to suppress the Brownian noise of the sense element
`to around this s,'une level for the sense element in Fig. 6.
`[33]. The more than two ord,rs of magnitude gap between
`measured performance and fundamental noise floor can be
`closed with improved CMOS device performance and low-
`noise circuit topologies.
`4.2. Angular accelerometer
`
`Fig. 7 is an SEM of the inerlial wheel of an angular accel-
`erometer that is sensitive to changes in angular rotation rate
`about the Z axis [36]. This design was fabricated in the first
`BiMEMS multi-project chip in 1994. The noise floor is 75
`rad s -2 Hz-J/2 which can be lowered by increasing the
`
`Fig. 6. Dic photograph of fully differenliaJ sigma-delta laleral accelerometer
`(upper left-hand comer chip in Fig. 5) [34,35].
`
`6
`
`
`
`R. ~. Howe ¢t ai. /Sensors and Act~tors A 56 f ! 996) 167-177
`
`173
`
`l
`
`nmqpm,~t,w
`Fig. 8 Block di~ram of the dual-input-axis (Xand ¥) vibratory rate IE~ro-
`scope [3~,3S1.
`
`metric vibration of the proof mass. The length of th:
`suspension is about 1 ram; the tensile residual stress in the
`BiMEMS structural polysilicon is unreleased in the clamped-
`clamped suspension and suppresses warpage due to the stress
`gradient. The initial design has a noise floor of 5" s- t in a 25
`Hz bandwidth, with active adjustment of the sense mode to
`wilhi, 5% of the driven mode at 12.5 kHz. With improve-
`mcnts in the se~,~e and control circuits and modifications to
`the sense element for additional sense capacitance, an order-
`of-magnitude reduction in thc noise floor is cxpccted [40].
`A final point about vibratory ratc gyroscopes is that thcy arc
`typically operated in vacuum in order to achievc mechanical
`quality factors greater than 1000, which both improves sen-
`sitivity and sim?lifics the sustaining amplifier design. Vac-
`uum micro-packaging at thc wafcr lcvcl is an attractivc
`solution to this requircmcnt [23].
`
`"",
`
`~! i!~
`
`:': ~"
`
`%
`
`Fig. 7. Scanning electron micrograph of the 460 t~na diameter inmliai wheel.
`used as the sense element in a Z-axis integrated angular accelemmeter fab-
`ricated in BiMEMS [36].
`
`inertia and also, by improved electronics [ 36]. In Fig. 7, the
`suspension design is largely constrained by the need to sup-
`p, ess out-of-plane warpage due to the residual stress gradient
`in the structural polysilicon layer.
`
`4.3. Vibratory rate gyroscopes
`Fig. 8 shows a block diagram of a dual-input-axis vibratory
`rate gyroscope [ 37,38]. The suspended rotor is sustained in
`resonance by means ofa trans-resistance amplificr with auto-
`matic gain control [ 39]. When the frame of reference of the
`chip is rotated about the X axis (horizontal in Fig. 8), the
`resulting Coriolis torque on the vibrating rotor causes it to
`oscillate about this axis. Underlying pie-shaped electrodes
`are used to detect oscillation of the sense mode. Since the
`structure is symmetric, it also detects angular rate about the
`Y axis. By means of adjusting the d.c. voltage on the sense
`electrodes, the frequency of the sense mode can be adjusted
`to match closely that of the driven angular mode (21 or 28
`kHz in two versions), which enhances the sensitivity of the
`gyroscope [38]. Experimental measurements for the 660/~tm
`diameter rotor shown in the die photograph in Fig. 9 .~how
`the noise floor in a 25 Hz bandwidth is 1.2 ° s- n without mode
`tuning and 10 ° min-t with tuning of the sense mode. The
`ultimate noise floor is a strong function of the rotor radius,
`which is limited by the stress gradient in the structural
`polysilieon.
`The vibratory rate gyroscope design sino~,m in Fig. 10 is
`sensitive to input rotations about t~¢ Z axis [40]. The proof
`mass is driven into resonance i:, the X axi~ and oscillates in
`the Yaxis due to the Y-directed Coriolis fc:~c resulting from
`rotation about the Z axis. Since no out-(,f-pla,¢ motion is
`involved in this gyroscope, closed-loop operation is feasible
`with a single structural polysilicon layer. An innovative fea-
`ture of this gyroscope, the sense element of which is shown
`in the die photograph in Fig. I i~ is the detection and electro-
`static suppression of the quadrature error signal due to asym-
`
`. . . .
`
`Fig. 9. D~ ~
`viba~ory ~ ~
`
`._
`
`_<"
`
`.
`
`.
`
`.
`
`.
`
`of tlw 660 ~m d~maxcr rowr fix dlc dual-inf.-axis
`fab~
`in BiMEMS [37,38).
`
`.
`
`.
`
`
`
`7
`
`
`
`174
`
`R.T. Howe et al. /Sensors and Actuators A 56 (1996) 167-177
`
`Fig. I 0. The sensing principle of the Z-axis vibratory rate gyroscope is shown at left. in which the Coriolis force in the Y-direction results from the cro~ product
`of the velocity of the driven X mode and the Z-axis input angular rate. The surface-micromachined sense element uses inner comb drives to sustain the resonant
`drive and outer interdigitated combs for sensing the Y-output motion [,10].
`
`E_~
`
`}vor
`rm
`
`• . balancing
`link
`
`pivot
`
`tether itii
`
`Fig. I !. Die photograph of the sense element for the Z-axis vibratory rate
`gyroscope, which is approximately 1 mmx I nun in area [40].
`
`4.4. Resonant accelerometer
`
`Fig. i 2 illustrates the operating principle of a differential
`resonant accelerometer fabricated in BiMEMS [41 ]. Comb-
`driven double-ended tuning forks are u3ed as the resonant
`sense elements. Their frequencies are shifted differentially
`by the axial compression and tension applied by the leveraged
`inertial force on the proof mass. The die photograph of a test
`sensor that uses lateral electrostatic forces to demonstrate the
`principle of operation is shown in Fig. 13. From measure-
`menus of the response to applied electrostatic forces, the pro-
`jec',ed sensitivity of a BiMEMS resonant accelerometer is
`about 5 Hz g- ~ for a 225 kHz nominal resonance frequency.
`The noise floor is determined by phase noise in the resonatols,
`which is dominated by electronic noise injected from the
`automatic gain-control loop through the delx nde,ce of fre-
`quency on oscillation amplitude. Resonant sensors have the
`
`Fig. 12. Schematic layout ofa surface-micromachined, differential resonant
`accelerometer [41 ].
`advantages of an inherently digital output and potentially
`wide dynamic range.
`In summary, there are a variety of sensor topologies that
`can be designed using a single-structural-layer polysilicon
`integrated microsystem technology. The promise of surface-
`micromachined sensors for addressing many real-world sens-
`ing needs, potentially in the same monolithic chip, has been
`demonstrated in the first [ 42 ] and second (Fig. 5) BiMEMS
`multi-project chip runs.
`
`5. Alternative microsystem technologies
`
`In addit:,on to polysilicon surface micromachining, several
`academic and industrial groups have merged other MEMS
`processes with CMOS technologies. Recent advances in dry-
`
`8
`
`
`
`R.T. Howe et aL / Sensors and Actuators A 56 ~ 19961 167-177
`
`i 75
`
`porate multiple isolated conductors [49], Finally, one limi-
`tation of integrated processes based on embedding MEMS
`into the subswate is the absence of a dense interconnectiou
`layer under:|eath the microstructure, corresponding to the first
`polysilicon layer in Fig. I.
`Integrated technologies based on metal microstractures are
`also competitive with polysilicon surface micromachining.
`Since the deposition temperatures axe much lower than for
`polysilicon, it is possible to achieve a modular process by
`completing the CMOS process before microstructure fabri-
`cation. The multi-structural level digital mirror display
`(DMD) process is intended for display applications but could
`also be applied to inertial sensors [51 ]. Plating is attractive
`for fabricating high-aspect-ratio structures and has been
`applied to fabricate the ring-resonant gyroscope after com-
`pletion of CMOS [ 52 ].
`
`6. Conclusions
`This paper has reviewed the technological challenges pre-
`sented by co-fabricating polysilicon surface-micromechani-
`cal su'uclmes and CMOS electronics. Modularity is attractive
`in a merged technology, especially with the need for dense
`CMOS and multi-structural level micromechanical struc-
`tures. The micromechanics-first approach using chemical
`mechanical polishing for planarization achieves this goal in
`a manufacu~mble technology [ 27 ].
`Remark.able progress has occurred over the past several
`years in extending the range of sensors that can be imple-
`mented in a first-generation polysilicon integrated microsys-
`terns technolo