Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Microsystems & Nanoengineeringhttps://doi.org/10.1038/s41378-019-0095-9 www.nature.com/micronano

ART ICLE Open Ac ce s s

Concept and proof for an all-silicon MEMS microspeaker utilizing air chambersBert Kaiser1, Sergiu Langa1,2, Lutz Ehrig1, Michael Stolz1, Hermann Schenk1, Holger Conrad 1, Harald Schenk1,2,Klaus Schimmanz2 and David Schuffenhauer1

AbstractMEMS-based micro speakers are attractive candidates as sound transducers for smart devices, particularly wearablesand hearables. For such devices, high sound pressure levels, low harmonic distortion and low power consumption arerequired for industrial, consumer and medical applications. The ability to integrate with microelectronic circuitry, aswell as scalable batch production to enable low unit costs, are the key factors benchmarking a technology. TheNanoscopic Electrostatic Drive based, novel micro speaker concept presented in this work essentially comprises in-plane, electrostatic bending actuators, and uses the chip volume rather than the its surface for sound generation. Wedescribe the principle, design, fabrication, and first characterization results. Various design options and governingequations are given and discussed. In a standard acoustical test setup (ear simulator), a MEMS micro speaker generateda sound pressure level of 69 dB at 500 Hz with a total harmonic distortion of 4.4%, thus proving the concept. Furtherpotential on sound pressure as well as linearity improvement is outlined. We expect that the described methods canbe used to enhance and design other MEMS devices and foster modeling and simulation approaches.

IntroductionThere is great demand for high performance micro

electromechanical systems (MEMS) micro speakers by theconsumer and by the hearing aid industry. Key factors,such as high sound pressure level (SPL), low total har-monic distortion (THD), low power consumption, smalldevice footprint and batch fabrication possibilities areneeded to outperform classic micro speaker technologieslike moving coil and balanced armature receiver tech-nology. MEMS-based micro speakers achieve all these keyfactors. Successful devices made with MEMS, like inertialsensors, timing devices and pressure sensors, are ubiqui-tous in many electronical applications. These applicationsrange from vehicles to wearables. A prominent example issmartphones, which have all the types of MEMS men-tioned above. They also include acoustic transducers,most likely appearing as multiple MEMS microphones, to

aid their original primary application, voice communica-tion. In contrast to fabrication, pricing and integrationlogic, today’s smartphones include fine mechanical engi-neered, non-MEMS micro speakers. Hearables, which arewearables plugged directly into the ear canal and relyingmostly on acoustics for their human-machine interface,are expected to take over the acoustic functionality ofsmartphones and extend their application to that of per-sonal assistant. In earbuds, the available space is limited,thus demanding exceptional miniaturization and inte-gration of all components including electronics, sensors,batteries, and acoustic transducers.Several concepts for MEMS-based micro speakers have

been reported in the literature. Among them are the fol-lowing examples. Cheng et al.1 published an electro-magnetically driven MEMS micro speaker comprising anelectroplated, micro machined membrane with a diameterof 3.5 mm and a permanent magnet assembled under-neath. The device reached 93 dB at 5 kHz in a 2 cm3

closed cavity, consuming 320mW of power. No resultsfor the THD were published. Chen et al.2 used a 3.5 mm

© The Author(s) 2019OpenAccessThis article is licensedunder aCreativeCommonsAttribution 4.0 International License,whichpermits use, sharing, adaptation, distribution and reproductionin any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if

changesweremade. The images or other third partymaterial in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to thematerial. Ifmaterial is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtainpermission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Correspondence: Bert Kaiser (bert.kaiser@ipms.fraunhofer.de)1Fraunhofer-Institute for Photonic Microsystems, 01109 Dresden, Germany2Chair of Micro and Nano Systems, Brandenburg University of TechnologyCottbus-Senftenberg, 03013 Cottbus, Germany

1234

5678

90():,;

1234

5678

90():,;

1234567890():,;

1234

5678

90():,;

diameter polymer membrane on a silicon substrate,incorporating a single loop copper coil and two magnetsfor electromagnetic driving of the speaker. They were ableto demonstrate 106 dB at 1 kHz in a 2 cm3 closed cavity,with an input power of 1.76 mW. No values for the THDwere reported. Albach et al.3 published the concept of aMEMS-based micro speaker driven by a changing mag-netic field. The concept describes a parallel arrangementof bending actuators, using a magnetostrictive active layerwith a photoresist passive layer. The actuators form anacoustically closed membrane of about 7.5 mm2, despitethe gaps between the individual cantilevers. In a hybridsetup, Albach et al. drove the MEMS devices with exter-nally generated magnetic fields. A SPL 101 dB was mea-sured in a 2 cm3 closed cavity at 400 Hz. No values foreither the THD or power consumption were presented. Asound transducer based on the piezoelectric bending ofactuators was presented by Stoppel et al.4. They describe aconcept where the laterally expanded, triangular actuatorsdeflect out-of-plane according to the bimorph effect. For adriving voltage of 2 Vpp they were able to measure a peakSPL of 86 dB (2 cm3 closed cavity) with a THD of less than2% including a preprocessed input signal. A sensitivity ofmore than 105 dB/mW was achieved. Roberts et al.5

presented the concept of an electrostatically actuatedMEMS micro speaker. It features a silicon carbide mem-brane on a silicon back-plate with an 8 µm electrostaticgap. They were able to measure a SPL of 73 dB at 16.6 kHzat a distance of 10 mm in free field radiation with a drivingvoltage of 200 Vpp. From the deflection and active areadata given, this equals an approximate SPL of 65 dB in aclosed 2 cm3 cavity. Neither the THD nor the powerconsumption data were reported. Fei et al.6 presented athermoacoustic device based on graphene foam. Thedevice exhibits an active area of 100mm2. An SPL of75 dB at 10 kHz in a free field measurement at 3 cm isreported for an input power of 1000 mW. This translatesto approximately 84 dB in a 2 cm3 cavity. For the sake ofcompleteness it should be noted that several works havebeen conducted with parametric transducers, utilized inarrays7, for digital sound reconstruction8–11. Theseapproaches, which are presented in literature, areassumed to be relatively large because of the large numberof single membrane drivers arranged in parallel. Thisrenders the devices not beneficial for miniaturization inthe first place.The literature overview reveals various mechanical

approaches, with at least three different actuatingmechanisms. Except for the thermoacoustic devices, theyall actuate single or multiple membranes or similarworking entities which deflect out-of-plane. Thesemembranes or plates exhibit different sizes and shapesand are sometimes closed or slitted.

Highly miniaturized MEMS concepts which do notrequire complex or elaborate assembly technologies, butrather allow the use of broadly available manufacturingtechnologies (are compatible with complementary metal-oxide semiconductor (CMOS) technology) are highlyfavored for scalable mass production.In this paper we present a novel design concept for

MEMS micro speakers based on the fabrication technol-ogies referred to as silicon bulk micro machining, which isfully CMOS compatible. The design comprises all-siliconelectrostatic, in-plane bending micro actuators working inair chambers, thus utilizing the chip’s bulk volume for airdisplacement and sound pressure generation. This microspeaker design approach allows the substitution of chiparea for volume. In this work governing principles andparameter relations are presented and the concept’spotential is estimated. By fabricating and characterizingfabricated specimen, the concept is proven. To validate aconcept, performance in terms of attainable sound pres-sure has to exceed the hearing threshold and should atleast reach sound pressure values of normal speech, e.g.,55 dB at 1 kHz, and surpass previously reported values forelectrostatic transducers5. Distortion in terms of THDshould be below 10% and more likely below 5% as indi-cated by listening test12.

ResultsThe nanoscopic electrostatic drive (NED) based micro

speaker concept is shown in Fig. 1a. The device comprisesclamped-clamped electrostatic bending actuators, theNED13, placed pairwise in rows and columns within thedevice layer of a bonded silicon on insulator (SOI) waferand covered by another wafer bonded to the SOI waferwith a tiny separation. Between each neighboring row ofactuators, acoustically effective openings are integrated inthe top and bottom of the wafer in an alternating way toallow the emission of sound from the device withoutacoustic shortcut.The clamped-clamped configuration, when based on a

suitable design, gives an efficient double-S bending shape tothe NEDs. Every other row of paired actuators is shiftedalong the longitudinal axis by half the length of one actuatorto make denser packing of the actuators possible whilstconsidering the bending shape. When a driving voltage isapplied across the electrodes of the individual but identicalactuators, they bend laterally in-plane according to theirorientation in an odd mode relative to each other, thusbalancing inertial forces within the device. With the top andbottom wafers as closing covers, each space between any ofthe actuators creates an air chamber. The function of eachchamber at any one moment is to push air out into or topull air in from the surroundings, depending on the direc-tion of its associated actuators’ active deflection.

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 2 of 11

As a consequence of the deflection of the actuators, afraction of the air volume ΔVact of a pushing chamber istransferred out of the device through the associatedacoustic opening. The same quantity of air is transferredinto the device, i.e., into the pulling chambers. When thedriving voltage is turned off, the restoring mechanicalforces of the actuator beam bring it back to its restposition and invert the previous volume flow. Therefore,each pushing chamber becomes a pulling chamber andvice versa. This process may be driven at any frequency fin the audible range, thus generating sound. Consideringinertial forces in a dynamic actuation regime the NEDsdeflection may also get inverted, i.e., deflection passesbeyond the mechanical rest position. A fact to be con-sidered in the device design. The actuators must not passthe acoustic opening aperture in any operational state, asthis will cause loss of sound pressure due to acousticshortcutting.The tiny separation between the in-plane deflecting

actuators and the top and bottom wafer ensures mova-bility of the actuators (see Fig. 1b). However, the separa-tion represents a fluidic path between oppositelypressured chambers, thus enabling an acoustic shortcut.Geometrically, the clearance should be designed in such away that losses are negligible, most likely by minimizingthe distance between the actuators and the covers. Incontrast, electrostatic forces generated by fringing elec-trostatic fields around the actuator tend to deflect theactuators in the z-direction, closing the clearance, and arereduced by a larger clearance. A vertical movement in thez-direction must be considered, since perfect symmetry ofthe upper and lower clearances is not to be expected inpractice. Since the vertical stiffness of an actuator isproportional to l-3, the size of the clearance limits theviable actuator length l. If not designed properly, thevertical movement ultimately results in pull-in if the

voltage exceeds the critical value Upi. This snap-in effect14

would create undesired friction between the actuator andthe bottom or cover wafers, resulting in distortion (“rub &buzz”15) and substantially deteriorated sound quality.Furthermore, vertical pull-in may also result in electricalshortcutting and may lead to device failure.The height of the actuators equals the SOI layer thick-

ness and specifies the displaced volume for a given lateralactuator displacement. In a closed acoustic emissioncavity, the sound pressure is proportional to the quotientof the volume change and the initial volume of the cavity.Thus, a given layout may be extended to an increasedheight into the SOI layer, thereby increasing the generablesound pressure without affecting the footprint of thedevice. This is a unique characteristic of the microspeaker concept presented in this paper and is in contrastto any other approach relying on membranes. The designpossibilities, some of which have already been describedabove, comprise actuator design, actuator arrangement,actuator count, fluidic channels and damping entities.These all further shape the uniqueness of the concept.Nevertheless, the various layout options result in a rathercomplex design process and must be closely aligned withthe technological boundary conditions, most likely relat-ing to minimal widths and lengths, to exploit the fullpotential of the concept. To estimate this potential, fun-damental aspects are presented in the following section.

Theoretical descriptionAttainable sound pressure is the prominent perfor-

mance indicator for micro speakers. In the followingsection, the limits are estimated using a geometricapproach. For a given SOI layer thickness h, the acousticperformance is then specified by the arrangement andnumber of actuators, each exhibiting a certain geometryand area consumption ANED. While being displaced, the

Airchamber

Deflected Undeflected

Device layerBottom wafer

h

ANED

Aact

z

yx

Acousticopening

Pair ofactuators

Separationlayers

Cover wafer

a bCover wafer Acoustic opening

Pulling chamberPushing chamberNED actuator

z

y

Separation/clearance

bcl lcl

lao

hao

h

Fluid flowActuator active movement

Fig. 1 MEMS micro loudspeaker utilizing air chambers. a schematic 3D representation. b Cross-sectional schematic (not to scale)

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 3 of 11

actuators sweep across an area Aact, thus defining the ratiorarea=Aact/ANED. The area Aact of the double-S bendingshape may be simply calculated from the actuators’maximum deflection at its middle yielding Aact=wmax * l/2. The area of the whole device Adev equals the sum of thearea of the actuators, considered a passive area, and theactively swept area, i.e., Adev=ANED+Aact=ANED * (1+rarea). Comparing the actively scanned area to the devicearea defines the ratio rfill=Aact/Adev= 1/(1+ 1/rarea).Essentially, rfill is in the range 0 < rfill < 1 and is practicallylimited to rfill << 1 because additional areas are requiredwithin Adev, such as for mechanical anchors and fluidic orelectrical paths, as well as areas related to technologicalconstraints. These areas may be subsumed into ANED forthe sake of simplicity.The SPL (symbol: Lp) in a closed acoustic emission

space of volume V0, such as with earphones plugged intothe ear canal, is not dependent on the position within thisvolume. Assuming wavelengths twice as big as the largestgeometric dimension of V0 and that V0 is without leakage,the SPL can be simply calculated as follows. First, thepressure increase Δp is calculated as:

Δp ¼ κp0ΔV=V0 ð1Þ

where κ, p0 and V0 are the adiabatic index, the initialpressure of air and the volume of the acoustic cavity,respectively. The total displaced volume ΔV is the sum ofall the volume changes caused by the single actuators,ΔVact. The SPL is then calculated as the logarithm of thepressure increase relative to an acknowledged referentialpressure pref:

Lp ¼ 20 logΔppref

14

ffiffiffi2

p� �ð2Þ

The sqrt(2)/4 factor accounts for the RMS calculation ofthe sound pressure, assuming a harmonic sinusoidal

movement of the actuators. Substituting ΔV=Aact * h=Adev * rfill * h gives

Lp ¼ 20 log κp0pref

AacthV0

14

ffiffiffi2

p� �Lp ¼ 20 log κp0

prefAdevrfillh

V0

14

ffiffiffi2

p� � ð3Þ

Obviously, the maximum SPL is attained when rfill= 1and when device area and SOI thicknesses are maximized.This implies that the ratio rarea has a practically useablelimit of about rarea= 10, giving an rfill of approximately 9/10. In other words, there is no interest in actuators whichexhibit a deflection greater than ten times their associatedlateral dimensions. A ratio rarea > 10 would not increaseattainable sound pressure significantly. Hence, the regimewith values of rarea < 10 is referred to as the design opti-mization regime. Values of rarea > 10 are in the saturationregime. Figure 2a plots the attainable SPL against rarea,showing the different regimes. The SPL is normalizedusing arbitrarily chosen parameters Adev, h and V0 so thatthe maximum achievable sound pressure for rfill= 1 andrarea≫ 100, respectively yields 100 dB, since it is a relativemeasure. Additionally, SPL may be estimated for differentmeasurement cavity volumes using Eq. (3).

The previous SPL calculation assumes that the fluidtransfer in and out of the device is through large enoughopenings which do not lead to, e.g., significant damping inthe audible frequency range. Another important aspect isthe split of fluid flows within the device, i.e., the differ-ences in flow resistance between different fluidic paths. Inthe following, a static analysis is given. The flow resistivityof the acoustic openings may be estimated from the law ofHagen-Poiseuille16:

_Vac ¼ K �min bao; haoð Þ3�max bao; haoð Þ12ηlao

Δpao ð4Þ

10−3 10−2 10−1 100 101 10240

50

60

70

80

90

100

110a b

Design optimization regime Saturation

Ratio rarea

Rel

ativ

e so

und

pres

sure

leve

l/dB

r

10−1 10075

80

85

90

95

100

Ratio uac/udc

Rel

ativ

e so

und

pres

sure

leve

l/dB

r

SPL

THD

0

5

10

15

20

25

Tot

al h

arm

onic

dis

tort

ion/

%

Fig. 2 Sound output dependencies on geometry and electrical driving. a Dependency of a relative sound pressure level (arbitrary reference) as alogarithmic measure versus design factor rarea. b Relative sound pressure level and THD (static domain) versus driving regime uac/udc for udc+ uac=const

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 4 of 11

where bao, hao and lao are the width, height and length ofthe acoustic opening channels respectively. It should benoted that the fluidic flow is in the direction parallel to thelength of the opening channel lao. The viscosity of air isencoded in ή. The correction factor Kmay be estimated asfollows16:

K ¼ 1�X1n¼1

1

2n� 1ð Þ5 �192π5

� min b; hð Þmax b; hð Þ tanh 2n� 1ð Þπ

2max b; hð Þmin b; hð Þ

� �

ð5Þ

The pressure amplitude Δpao may be derived from Eq.(1), taking V0 as the volume of one pushing or pullingchamber within the device (refer to Fig. 1b) and sub-stituting ΔV with two single actuator displaced volumesΔVact, so that Δpao ∝ 2 ΔVact. The fluidic losses throughthe clearance, the rectangular opening along the actuator,may also be estimated. Two clearances of similar geo-metry are active for each actuator, with the Δp beingdoubled because they have opposite signs on either side ofthe actuator, giving Δpcl ∝ 2 ΔVact. In contrast to theacoustic openings, a high fluidic flow resistance is desiredfor the clearance channels. The clearance channel haslength lcl, width bcl and height hcl. According to a typicallayout, it is assumed that bao= bcl, hao≫ hcl and lao≫ lclholds true. The ratio between the flow resistances of theclearance and acoustic openings may then be simplified asfollows:

_Vcl

_Vao/ 2

h3clh3ao

� laolcl

� 1 ð6Þ

Thus, the fluidic flow around the actuators through theclearance is significantly lower than through the acousticopenings. Consequently, no reduction in the acousticallyeffective volume flow must be assumed because of the airchamber concept.The SPL described before defines the objective loudness

of the acoustic transducer, i.e., acoustic quantity. Incontrast, the THD defines the acoustic finesse, i.e., theacoustic quality of an acoustic transducer, making it animportant measure for comparing sound reproduction.The THD shall therefore be discussed briefly and theo-retically estimated in the following.The working principle of the NED transducer is based

on Coulomb forces with a quadratic dependence on thedriving voltage as well as on the inverse of the electrodedistance (Coulomb nonlinearity), which diminishes withincreasing voltage13. The latter aspect is greatly influencedby the mechanics defined through the layout and how thisaffects the electrodes’, and consequently the actuators’,movements. For a clamped-clamped actuator configura-tion as given in our approach, a stress, resulting fromdeflection and inducing stiffening, i.e., nonlinearmechanical behavior known as the Duffing effect17, is tobe expected. This effect develops contrarily to the

previously mentioned electrode separation nonlinearitywith increasing deflection of the whole actuator. There-fore, the focus is on the quadratic proportionality on thedriving voltage only (a simplification). Adding an offsetvoltage udc is commonly known to enhance linearity, i.e.,to decrease the distortion. Thus, the driving voltage is:

U ¼ uac cos ωtð Þ þ udc ð7ÞThe THD (symbol k; according to International Elec-

trotechnical Commission IEC) is defined as the sum of allpower pi radiated in frequencies other than the desiredexcitation frequency, relative to the total emitted soundpower.

k ¼ 100%

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPni¼2 piPni¼1 pi

sð8Þ

The actuator force F and deflection x, and thus its soundgeneration capability, are proportional to the square of thedriving voltage U:

F tð Þ � uac cos ωtð Þ þ udcð Þ2¼ u2ac2

þ u2dc

� �

þ2uacudc cosðωtÞ þ u2ac2

cosð2ωtÞð9Þ

The average radiated power pi is then proportional tothe square of the signal:

P / 1T

ZT0

dt x2ðtÞ with x / F ð10Þ

Comparing the amplitudes in Eq. (9), k reveals itsdependency as follows:

k ¼ 100%

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiu2ac

u2ac þ 16u2dc

swith juac þ udcj<Upi ð11Þ

Because of the voltage limits due to pull-in, the sum ofoffset voltage and signal voltage need to be smaller thanthe pull-in voltage Upi at any time. Therefore, the ratiouac/udc needs to be minimized to minimize the THD. TheSPL is dominated by the proportionality of udc and uac.The SPL is then maximal for uac/udc= 1. In Fig. 2b, SPLand THD are plotted against the ratio uac/udc. From this, aratio uac/udc < 1, e.g., uac/udc= 1/8, seems reasonable toget low distortion of around 3% and a relative SPL of−8 dB (factor 0.4). For real devices, the SPL and THDfollow more complex dependencies, including themechanical nonlinearities mentioned above and fluidicinteractions, e.g. from the acoustic load.

Transducer designA first design was created and fabricated. The layout of

the SOI layer is shown in Fig. 3a. It shows an arrangement

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 5 of 11

of 14 pairs of long actuators in a 2-serial-7-parallel (2s7p)configuration, meaning that there are two actuator pairsin a row, and seven rows aligned parallel to the deviceplane. The actuators are bundled into three individuallyaddressable groups (3-1-3 rows). To increase the fill factorof the design, remaining space to the edge of the preferredrectangular shaped device is filled with actuators of halfthe length. This adds seven pairs of half-length actuatorsin total. The footprint of the rectangular active area is9.3 mm2. Each of the double clamped actuators has alength of 2.2 mm and 1.1 mm respectively. Vertical pull-infor the maximum driving voltage of 48 V is safely avoided,even after taking fabrication variations into account. TheNED actuators’ shape was designed with finite elementanalysis-based optimization with the figure of merit beingthe displacement achievable while respecting technologi-cal boundary conditions (electrode thicknesses, trenchwidths, material properties). Details of the actuatorarrangement and design properties are shown in Fig. 3b, c.The approximate width of the actuators is less than

20 µm. Each pair of actuators exhibits a distance of about84 µm while each row of actuators is placed at about110 µm from each other.The device layout incorporates large (250 × 250 µm2)

bond pads for wire bonding afterwards. Acoustic openingsin the bottom and cover wafers are designed as rectangleslocated between every other actuator pair. The size of theopening is determined by the technological boundaries foretching, which defines their minimal width (30 µm). Thebottom and top wafers are 400 µm thick. The distance(clearance) between the actuator and the bottom waferequals the thickness of the 1 µm buried oxide layer of thebonded SOI wafer. This is also the design value for theseparation to the top wafer, which can be defined by adeposited and patterned bonding interface layer. The SOIlayer thickness is 75 µm.Three different transducer devices were fabricated on

one single chip. Fabrication was performed on siliconwafers, solely using CMOS compatible processes. Themain process is deep reactive ion etching (DRIE). Details

Bondpadsa

bc

Pair of short actuatorsPair of long actuators

Mechanical anchor

First actuator column Second actuator columnActive area

500 µm

Acoustic opening (bottom wafer)

550 µm

50 µm

84 µm

275 µm

80 µm

110 µm

40 µm20 µm3 µm

2.5 µm

Fig. 3 Geometric setup of fabricated MEMS speaker a Top view of the device layer showing actuators and their arrangement. Sample was takenout during fabrication without the cover wafer being bonded. b Enlarged partial view of three rows of actuators. c Detail of an actuator includingdimensions

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 6 of 11

of the fabrication are supplied in the methods andmaterials section.

Measurement setup and resultsTo characterize the acoustic performance, SPL and

harmonic distortion tests were performed. The bare

MEMS die layer stack, comprising the actual device, thebottom wafer and the cover wafer bonded on top, wasglued to a carrier board which allows for electrical con-tacts using wire bonding (Fig. 4a). The layer stack isshown in Fig. 4b. A specially designed adaptor embracingthe carrier board and the MEMS device allowed for the

1 mm

a

400 µm

400 µm

75 µm

Cover wafer

Device wafer

Bottom wafer

b

102 103 104 102 103 10460

70

80

90

100

110

Frequency/Hz

Sou

nd p

ress

ure

leve

l/dB

MeasurementModel: transducer/ear simulatorModel: transducer/pure cavity

0

10

20

30

40

Frequency/Hz

Tot

al h

arm

onic

dis

tort

ion/

%

MeasurementModel: transducer/ear simulatorModel: transducer/pure cavity

c d

50 µm 50 µm 50 µm

e

Fig. 4 MEMS device setup and acoustical as well as optical measurement data a Assembled device with cover wafer on top showing itsacoustic openings; the MEMS chip is glued to a carrier board enabling wire bonds for electrical contacting. The chip actually comprises three differentmicro speakers, which were not diced for this test set-up. The middle one was used as shown in (a). b Side view of the bonded wafer stack showingthe three wafers and their thicknesses. Scallops resulting from cleavage through lacer dicing are clearly visible. c Measurement and modeling resultsfor the SPL. Device was driven with udc= 40 V and uac= 10 Vpp. An SPL of 69 dB was achieved at 500 Hz. d THD measurement and modeling. THDvalue at 500 Hz is 4.4%. e Sequence of stroboscopic pictures from the holographic microscope showing three consecutive (time wise) deflectionsstates in resonance of one actuator pair (specimen without cover wafer)

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 7 of 11

mounting of a standardized IEC 60318-4 ear simulator(formerly IEC 60711,18) incorporating a condensermicrophone measuring in a closed cavity of 1.26 cm3 (atthe back end of the chip). Additionally, the adaptorensured decoupling of the acoustic openings in the topand bottom wafers by framing the MEMS device. Thefront side of the chip was left open to the surroundings.Using an audio analyzer (NTI Flexus 100) and a voltageamplifier (Krohn-Hite Model 7602M) to drive the device,the acoustic response to sinusoidal excitation (udc= 40 V,uac= 10 Vpp) while sweeping the frequency was obtained,as shown in Fig. 4c, d. In addition, these figures also showthe simulation results of the MEMS transducer and earsimulator combined, as well as of the MEMS transducerwith an acoustic output impedance as specified in ref. 19.Details of the simulation models are supplied in the“Materials and methods” section.For frequencies up to 1 kHz, the SPL measurement (pref

= 20 µPa) shows a comparatively slight variation between68 and 70 dB. In this frequency range, the quasistaticbehavior of the MEMS transducer driving the acousticload is visible. At a referential frequency of 500 Hz, theSPL was measured at 69 dB without being influenced bydynamical effects of either the transducer itself or the earsimulator. Beyond 1 kHz, these dynamical effects becomevisible, resulting in an increase in the SPL. Two char-acteristic frequencies may be distinguished, 1.7 kHz and11 kHz. Both are attributed to the IEC ear simulatorcharacteristics and may be estimated from equivalentcircuit models (see methods and materials section). Amaximum SPL of 104 dB was measured at 11.4 kHz. Thisis a superposition of the ear simulator resonance and thefirst mode transducer resonance. This first mode trans-ducer resonance is also visible as a shoulder at a frequencyof 9.1 kHz, agreeing well with electromechanical mea-surements of the long actuators. The contribution tosound generation from the short actuators may beneglected due to their low number, short length (cumu-lative length ratio 1/4) and very low deflection (x ∝ l−2).From the curve of the transducer with a pure cavity, it canbe concluded that the system is quite strongly damped(quality factor approximately Q= 2). This curve shows asignificantly higher SPL than the measurement for fre-quencies below 5 kHz. The losses in the measurementsare attributed to dissipation within the ear simulator (i.e.,its RCL-network).The THD remains below 5% for frequencies up to

500 Hz, representing the quasistatic behavior of thetransducer itself. An increase to <7% at half the 1.7 kHzresonance frequency of the ear simulator can be seen. TheTHD exhibits maximum values at a frequency of 5.7 kHz,the equivalent of the resonance of the ear simulator at11.4 kHz as mentioned before. Again, the signature of thetransducer itself is visible as a shoulder in the curve at

around 4.6 kHz, i.e., half its own resonance frequency. Incontrast, the THD drops at the resonances of the trans-ducer and the ear simulator. As before, the curve of thetransducer with a pure cavity load exhibits no specificpeaks except for the maximum at half the actuator reso-nance frequency.For comparison, pure electromechanical measurements

have also been carried out using a digital holographicmicroscope (Lyncee Tec R-2203). By taking a specimenwithout cover wafer (see Fig. 3) the lateral deflectioncould be obtained as a function of the driving signal fre-quency. The measurement results reveal a quasi-staticdeflection of about 1.5 µm at the middle of the 2.2 mmlong actuators for udc= 40 V and uac= 10 Vpp (f=50 Hz). These values fit well with the simulated deflectionof the clamped-clamped beams. At resonance a deflectionof about 22 µm was measured. Figure 4e shows a timewise sequence of consecutive deflections of oneactuator pair.The device’s total electrical capacitance was measured

as 65 pF with an impedance spectrometer (KeysightE4990A) at a frequency of 500 Hz according to DIN60268-7. Internal losses and dissipative effects, such ashysteresis losses, eddy current losses and heat generation,may be neglected for electrostatic transducers20. Assum-ing an efficiency of 40% for an appropriate driving circuit,handling the mostly reactive load of the electrostatictransducer, this gives a calculated sensitivity of 100 dB/mW (500 Hz, 1 mW, IEC 60318-4 ear simulator).

DiscussionThe fabricated and characterized device had a max-

imum SPL of 69 dB at 500 Hz in a IEC-ear simulator witha significant loss of SPL after resonance at around 10 kHz.A commonly known referential frequency response ofheadphones is the “Harman In-Ear Target ResponseCurve”21,22, which is the result of extensive listening teststo identify the preferred frequency responses of head-phones. It shows a steep decline of the SPL from 9 kHzongoing. A similar qualitative trend may be identified inthe measurement curve. Thus, the applicability of thedevice does not seem to be limited by the specific reso-nance frequency of the actuators.The measured SPL is enough to validate the new

loudspeaker concept as an audible sound reproductiondevice. The THD for the transducer shows values ofaround 4.5% without any signal preprocessing at all. Fromthe very simple theoretical estimations for the drivingregime, one would expect a THD of about 3%. The morecomplex and still purely linear equivalent circuit modelconsiders several more effects, such as the frequencydependencies of the clearance and the deflection. It isthen able to predict the behavior in good agreement withthe measurements for frequencies below any resonance.

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 8 of 11

To lower the THD values in general, i.e., below 4.4%,design optimization is proposed. This incorporates theuse of mechanical and fluidic nonlinearities counteractingthe inherent Coulomb nonlinearities. Additionally, settingup the transducer for its acoustic load, i.e., the earsimulator, e.g., by diversifying the resonances, will help toenhance the audio quality further.Based on the measured values of the quasistatic

deflection (1.5 µm) of the double-S bending shape, anactively swept area of approximately Aact= 1650 µm2 peractuator (28 in total) may be calculated. The estimatedsound pressure then yields 77 dB in a 1.26 cm3 volume.This value arises from the geometric approach presentedin the theory section neglecting any loss in the RCLnetwork of the ear simulator. Consequently, the value isgreater than the measured SPL result (69 dB) in the earsimulator with the same volume.Assuming a width of 20 µm of the actuator (Aned=

20 µm * 2.2 mm), a factor rarea= 3/80 may be estimated.This calculation neglects the additional required areas(e.g., bondpads, technology constraints, etc.) which wouldlower the factor because the effective Aned would increase.As such, the very same factor calculated from the soundpressure yields rarea= 1/500 (rfill= 1/500). It follows thatthe present proof-of-concept is in the design optimizationregime. For the boundaries (chip area of about 9.3 mm2

and SOI thickness of 75 µm), the maximum theoreticalSPL with the presented concept is roughly 123 dB in a1.26 cm3 cavity. To paraphrase, a rather large improve-ment of 54 dB (123–69) can be expected theoretically.Practically, an improvement to the design, mostly bydenser packing of more effective actuators, i.e., anincrease in rarea, will allow an increase in rfill to > 1/20,thus potentially adding > 20 dB.The work presented shows a MEMS micro speaker

concept and its proof with a fabricated specimen. Severalimportant advantages of the concept comprise CMOScompatibility including integration capabilities with driv-ing circuitry, precision mass production capabilities,possible low area foot print for high sound quality andquantity and very low power consumption.

OutlookFurther work focusing on SPL increase and THD

decrease is being conducted. Denser packing of theactuators thanks to design optimization is expected toallow higher efficiencies, i.e., higher SPL per area foot-print. The nature of the concept allows the volume of theMEMS die to be used for sound generation. This cap-ability is extended significantly if the thicknesses of thedevice layer is increased. Theoretically, the layer thicknesscan be increased from 75 µm to values as high as 725 µm,a typical silicon wafer thickness. This would add 20 dB forthe same area footprint and fill factor. This scaling is

connected with significant challenges in the technology,most prominently the DRIE process, which limits theachievable aspect ratio. The concept also allows for lateralacoustic openings, thus enabling the stacking of multipleMEMS dies to increase the SPL per area. Further opti-mization potential regarding distortion is expected to bereleased, with an optimized mechanical layout whichmakes use of the flexural behavior due to theactuators shape.Future work will also address the speaker system, i.e.,

the development of driving circuitry comprising a step-upconverter, signal amplifiers and, optionally, signal pro-cessing specifically designed for the device’s needs. Thechallenge is to maintain the advantages of the conceptregarding ultra-low power consumption. Recently, Häns-ler et al. have published work on capacitive loads andhighly efficient driving circuitry23. They describe severalamplifier classes as well as an approach to regain energyfrom reactive power. This is also of significant importanceto the capacitive driven speaker concept presented here.

Materials and methodsFabricationThe MEMS micro speaker was fabricated using

parylene-assisted wafer bonding, DRIE etching andatomic layer deposition (ALD). The process flow is illu-strated in Fig. 5a. A 200mm diameter, highly boron doped(1018 cm−3) bonded SOI wafer with a 75 µm device layerthickness was used as a substrate. First, open trencheswere etched and subsequently filled with alumina usingALD, forming the spacers as mechanical connectors butelectrical insulators in the NED actuators. Next, opentrenches were etched, using DRIE again to build theactuators, anchors and the air chambers. Then, acousticopenings were etched from the backside through thehandle layer (400 µm) of the bonded SOI wafer. The coverwafer was prepared by depositing and patterning the 1 µmoxide layer facing the device layer and by etching theacoustic openings. Finally, the cover wafer (400 µm) wasbonded on top of the device wafer using parylene (500 nmdeposited on the cover wafer, 300 °C) to assist waferbonding24.

Equivalent circuit modelingThe acoustic measurement results show the response of

the MEMS transducer to the load of the IEC ear simu-lator. In this section, both the transducer and the IEC earsimulator are modeled using an equivalent acousticcircuit.The equivalent circuit for the NED actuators is based on

a simple plate-capacitor model presented in the litera-ture25, which is transferred to an equivalent circuitaccording to the principles presented in26. The equivalentcircuit model of the NED-based drive of the micro

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 9 of 11

speaker comprises the electrostatic, mechanical andacoustic domains. The model is shown in Fig. 5b. In thismodel, the NED effect is implemented by a leveragingfactor translating the plate displacement to the beamdisplacement, and is encoded in TEM. The actuator massm is estimated from the layout. The electrical capacitanceC0 and the resonance frequency were taken from mea-surements. The resonance frequency of 9.1 kHz wasestimated from electromechanical measurements (LynceeTec R-2203) of the transducer without a cover wafer,taking into account the shift due to the offset voltageudc= 40 V. Effective stiffness k0 is then calculated fromresonance frequency and mass. The model is fitted usingthe pull-In voltage, leveraging factor (TEM) and dampingcoefficient r as fitting parameters. Pull-In voltage wasfitted to 49 V as it agree with the simulation. The lever-aging factor was also extracted from simulation in which avalue of 51 was calculated. The difference to the fittedvalue of 42 is attributed to the rather complex mechanics.The damping has a substantial influence on the fitting andthe fitted damping coefficient correlates to a quality factorof about 2.3. This value was qualified by acoustic mea-surements yielding a value of 1.5. For comparison, a

quality factor of approximately 30 was calculated from themechanical measurements in ambient atmosphere. Theobserved large damping in the devices with covers isattributed to the fluid dynamics of the vents and theinvolved flow pattern inside the 28 actuator chambers.The equivalent circuit for the IEC ear simulator is

derived from the literature18, and uses the parametriza-tion published by Brüel and Kjær19. The pure cavitycondition reflects all but one component removed fromthe ear simulator network. The remaining component isthe capacitor as shown in Fig. 5c right. The value of thecapacity is 2.1 pF, which is equivalent to a 297 cm3 volumeof a cavity.The models of both the transducer and the ear simu-

lator are used to estimate the SPL and THD. The couplingof both models takes into account a certain amount ofleakage between the ear simulator and the MEMS device.For the curves of the transducer with a pure cavity, pre-sented in Fig. 4b, c, the most part of the IEC ear simulatorcircuit is removed. Consequently, these curves revealbehavior dominated by the MEMS transducer. Themodels will be presented in greater detail as well as withfurther development in a forthcoming publication.

# 1 Etching of opentrenches and fillingwith alumina

Si SiOx Al2O3

#2 Etching of opentrenches for gapand movement

#3 Passivationand openingof backside

#4 Etching openingin bottom wafer

#5 Release etching andremoval of protectiveand sacrificial layer

#6 Parylene-assistedbonding of coverwafer with etchedacoustic openings

US c0C0

ZS

ZC F1

m 1k0

r

F2 P

1: TMA1 : TEM

−c0C0IS v1 v2 q

a

bEar simulator Pure cavity

IEC 60318-4P P C

c

Fig. 5 Fabrication of the device and equivalent circuit model. a Schematic process flow for the fabrication of the micro loudspeaker. The NEDactuator shows an arrangement of three electrodes, each separated by an electrode gap. The electrodes are separated electrically by spacers made ofalumina (ref. 13). b Equivalent circuit model of the micro speaker based on electrostatic bending actuators (NED). c Acoustic loads used. The earsimulator is taken from the literature18. The pure cavity condition represents a capacitor of 2.1 pF equivalent to a 297 mm3 volume

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 10 of 11

AcknowledgementsThis work was funded by the German Federal Ministry of Education andResearch (BMBF) as part of the MEMSound project in the funding measure“Validation of the technological and social innovation potential of scientificresearch - VIP+” under the grant number 03VP01800.

Authors’ contributionsB.K. created the layout of the MEMS device, accompanied fabrication andcoordinated characterization. S.L. was responsible for fabrication in clean roomfacilities and consulted on layout. L.E. conducted acoustic setup andcharacterization and co-created equivalent circuit models. M.S. performedmechanical and electrical measurements. H.S. proposed and developed theequivalent circuit models. H.C. and H.S. proposed the concept and revised themanuscript. K.S. proposed and discussed the governing modeling approaches.D.S. supported the optimization of the NED actuators.

Conflict of interestThe authors declare that they have no conflict of interest.

Received: 27 February 2019 Revised: 7 August 2019 Accepted: 12 August2019

References1. Cheng, M.-C., Huang, W.-S. & Huang, S. R.-S. A silicon microspeaker for hearing.

J. Micromech. Microeng. 14, 859–866 (2004). https://doi.org/10.1088/0960-1317/14/7/004.

2. Chen, Y. C. & Cheng, Y. T. A low-power milliwatt electromagnetic microspeakerusing a PDMS membrane for hearing aids application. 1213–1216 (2011).

3. Albach, T. S., Horn, P., Sutor, A. & Lerch, R. Sound Generation Using a Mag-netostrictive Micro Actuator. J. Appl. Phys. 109(7), https://doi.org/10.1063/1.3554200 (2011).

4. Stoppel, F., Männchen, A. & F. Niekiel, F. New integrated full-range memsspeaker for in-ear applications. MEMS (Belfast, 2018).

5. Roberts, R. C. et al. Electrostatically Driven Touch-Mode Poly-SiC Microspeaker.IEEE Sensors (2007).

6. Fei, W., Zhou, J. & Guo, W. Low-voltage Driven Graphene Foam ThermoacousticSpeaker. Communications, 1068–1071 (Belfast, Northern Ireland, UK, 2018).

7. Glacer, C., Dehé, A., Nawaz, M. & Laur, R. Reversible acoustical transducers inMEMS technology (DTIP 16–18 April 2013, Barcelona, Spain, 2013).

8. Diamond, B., Neumann, J. & Gabriel, K. Digital sound reconstruction using arraysof CMOS-MEMS microspeakers, https://doi.org/10.1109/MEMSYS.2002.984260,(2002).

9. Pedersen, B. R. Performance analysis of digital loudspeaker arrays. J. Audio Eng.Soc. 56(No. 9), 2008 (2008).

10. Dejaegera, R. et al. Development and Characterization of a PiezoelectricallyActuated MEMS Digital Loudspeaker, Proc. Eurosensors XXVI, September 9–12,2012 (Kraków, Poland, 2012).

11. Arevalo, A., Castro, D., Conchouso, D. Kosel, J. & Foulds, I. G. MEMS digitalparametric loudspeaker, MEMS 2016, Shanghai, China, 24–28 January 2016(2016).

12. Temme, S., Olive, S., Tatarunis, S., Welti, T. & McMulli, E. The CorrelationBetween Distortion Audibility and Listener Preference in Headphones, AudioEngineering Society, 137th Convention, 1–15 (Los Angeles, USA, 2014).

13. Conrad, H. et al. A small-gap electrostatic micro-actuator for large deflections.Nat. Commun. (2015). https://www.nature.com/articles/ncomms10078, https://doi.org/10.1038/ncomms10078.

14. Zhang, W.-M., Yan, H., Peng, Z.-K. & Meng, G. Electrostatic pull-in instability inMEMS/NEMS: A review. Sens. Actuators A214, 187–218 (2014).

15. Klippel, W., & Seidel, U. Measurement of Impulsive Distortion, Rub and Buzzand other Disturbances (AES, 2003).

16. Bruus, H. Theoretical Microfluidics (Oxford University Press, 2008).17. Brennana, M. J., Kovacicb, I., Carrellaa, A. & Watersa, T. On the jump-up and

jump-down frequencies of the Duffing oscillator. J. Sound. Vibration 318(4–5),1250–1261 (2008).

18. International Electrotechnical Commision, Electroacoustics—simulators ofhuman head and ear—Part 4: Occluded ear simulator for the measurement ofearphones coupled to the ear by inserts (International ElectrotechnicalCommission, Geneva, Switzerland, 1981).

19. Brüel & Kjær. Application Note Ear Simulator for Telephonometry (Brüel&KjærSound & Vibration Measurement A/S, Denmark, 2012).

20. Hunt, F. V. Electroacoustics - The Analysis of transduction, and Its His-torical Background. (Harvard University Press, Cambridge, Massachu-setts, 1956).

21. Olive, S., Welti, T. & Khonsaripour, O. The Influence of Program on SoundQuality Ratings of In-Ear Headphones, 142nd Audio Eng. Soc. Convention(Berlin, 2017).

22. Olive, S., Welti, T. & Khonsaripour, O. The Preferred Low Frequency Response ofIn-Ear Headphones, AES International Conference on Headphone Technology,paper 6-1 (2016).

23. Hänsler, M., Strle, D. & Auer, M. Design of a MEMS speaker driver. Elektro. Inf.135/1, 10–17 (2018).

24. Kim, H., & Najafi, K. Characterization of low-temperature wafer bonding usingthin-film parylene, J. Microelectromech. Syst. 14, 1347–1355 (2005).

25. Pelesko, J. A. & Bernstein, D. H.Modeling MEMS and NEMS (CHAPMAN & HALL/CRC, 2003).

26. Beraneck, L. L. Acoustics (Acoustical Society of America, Cambridge, Massa-chusetts, 1993).

Kaiser et al. Microsystems & Nanoengineering (2019) 5:43 Page 11 of 11