5386 IEEE SENSORS JOURNAL, VOL. 18, NO.

13, JULY 1, 2018

MEMS Capacitive Microphones: Acoustical,

Electrical, and Hidden Thermal-Related Issues
Germano Nicollini and Daniele Devecchi

Abstract— This paper reports on the design and implemen-

tation of high performance MEMS microphones. The main
parameters defining microphone performance, i.e., sensitivity,
acoustic overload level, and signal-to-noise ratio, have been
extracted from an acoustical–mechanical–electrical model of the
MEMS sensor and designing a proper architecture of the readout
circuit. Moreover, some subtle problems linked to the integration
of the MEMS sensor with its readout circuit in the same
package have been highlighted. These hidden issues are the
degradation of the power supply rejection at low frequencies,
the contamination from radio frequency interference, and the
presence of a spurious tone in the audio band when digital
microphones are operated in stereo configuration. They can
seriously impact the performance of advanced analog and digital
MEMS microphones if not properly tackled and solved. The
MEMS are made by a polysilicon membrane with serpentine
spring structures placed at the corners of the diaphragm, whereas
the readout application-specified integrated circuits are realized
in a 0.5-µm single-poly, four metal layers, N-well CMOS process
with MIM capacitors.
Index Terms— MEMS, sensors, analog and digital
microphones, sensitivity, signal-to-noise ratio, noise, acoustic
overload level conducted and radiate interference, power-supply
rejection, acoustics.

I. I NTRODUCTION

C ONDENSER MEMS microphones are becoming more

and more common in applications such as mobile
phones, tablets, laptops, PDAs, and hearing aids [1] because of
Fig. 1. Block diagram a) and cross-sectional view b) of a capacitive
MEMS microphone.

their advantage of packaging in surface mounting devices that

can withstand the standard lead free surface mount soldering. in the range of 10V-20V to bias the sensor, mandating the
Moreover, they have durable and reliable characteristics in inclusion of a charge-pump circuit and a high-voltage process
mechanical and electrical performance of the silicon material option in the readout circuit.
at high temperature and humidity conditions, and they can In reality there are some more problems, linked to the
be made in batch fabrication at low cost. Recent applica- integration of the sensor with its readout circuit in the
tions demand increased performance of the MEMS micro- same package, which, if not tackled properly, can seriously
phone in Signal-to-noise Ratio (SNR) and Acoustic Overload impact the performance of advanced analog and digital MEMS
Level (AOL). Since the SNR is linked to the lowest sound microphones.
level that the microphone can recognize above the noise floor, In Section II, the architecture and expected performance of
a higher SNR can extend the distance between the sound the realized MEMS microphone are presented by exploiting
source and the microphone, and also improve the quality of the an equivalent electrical circuit representing the mechanical
captured signal. A higher AOL allows increasing the dynamic and acoustical parts of the transducer. Section III deals with
range while conserving sound quality. The only reported the experiments that put in evidence three “hidden” issues
drawback of MEMS microphones is the need of a high voltage in MEMS microphones, whereas their origins are explained
in Section IV. In Section V, solutions supported by experimen-
Manuscript received April 26, 2018; accepted May 8, 2018. Date of tal results are presented. Finally, Section VI gives conclusions.
publication May 16, 2018; date of current version June 12, 2018. The associate
editor coordinating the review of this paper and approving it for publication
was Prof. Vedran Bilas. (Corresponding author: Germano Nicollini.) II. MEMS M ICROPHONE A RCHITECTURE
The authors are with STMicroelectronics, 20010 Cornaredo, Italy (e-mail:
[email protected]). A block diagram and a cross-sectional view of the MEMS
Digital Object Identifier 10.1109/JSEN.2018.2837143 microphone are shown in Fig.1.
1558-1748 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 NICOLLINI AND DEVECCHI: MEMS CAPACITIVE MICROPHONES 5387

Fig. 3. Expanded cross-sectional sensor view with main parameter definition.

Diaphragm impedance represents the movable membrane with

Fig. 2. Electrical circuit for the transducer’s mechanical and acoustical parts.
its supporting springs. L d accounts for the inertial mass of
the diaphragm, whereas Csp refers to the total compliance
They are basically made by a capacitive transducer, that of the springs. Ventilation holes are thin channels between
converts the incoming pressure variation into a voltage, and the substrate and the diaphragm that play the role of pressure
a readout circuit bonded together and mounted on a common equalization or compensation on both sides of the membrane.
substrate inside the same package. The substrate contains a Since they represent an acoustical short circuit, sizing of
small PCB that is used to route signals from the readout IC these channels is very important. If their impedance is too
to the pins outside of the package. In the case of analog low, most of the air flows through them and only a small
microphones (inside the green box), the readout circuit consists fraction of the air makes the membrane to move, reducing the
of a high input impedance low-noise preamplifier that drives transducer sensitivity. Otherwise, if it is too high, it can be a
an external load. In the case of digital microphones (inside the strong noise source, as we will see later on. The impedance
red box), the preamplifier is followed by a 1b  converter of these channels Rv accounts for the losses due to viscous
to perform a digitalization of the output voltage [2], [3]. resistance of the air passing through the ventilation holes,
Alternatively, a multibit  modulator followed by a noise- and L v represents the inertial effect of the air mass in the
shaper [4] can be employed. The sensor biasing voltage is channel. Since in the audio band Rv is dominant, L v has been
generated by a charge-pump circuit in the readout chip and is omitted in Fig. 2. Air gap impedance represents the flow of air
set by a huge resistance RBIAS such that the cutoff frequency in the gap between the moving membrane and the fixed back-
fB = 1/(2πRBIAS CSENSOR ) is well below the audio band plate. Ra accounts for the air squeezing inside the air gap when
(20Hz-20kHz). Since for frequencies above 20Hz the charge in the diaphragm vibrates, whereas L a represents the inertial
the capacitive transducer is constant, sound pressure changes effect of the air mass in the channel. Acoustic hole or back-
are translated into output voltage variations. plate impedance represents the air flow through the holes in
the fixed back-plate. Rbp accounts for the viscous dissipation
A. MEMS Sensor: Electro-Mechanical-Acoustic Analogy of the air streaming through the acoustic holes, whereas L bp
represents the inertial effect of the air mass in the acoustic
An equivalent electrical circuit of the sensor can be obtained
holes. Back-chamber impedance represents the interaction of
by relying to an electro-mechanical analogy [5] where a
the back-chamber with the sound. The back-chamber prevents
force F corresponds to a voltage and a velocity v plays the
external sound from driving the backside of the diaphragm
same role of a current, and to an electro-acoustical analogy
since having a closed backside can result in an air spring that
where the sound pressure P corresponds to a voltage and the
stiffens the diaphgram. Cbc accounts for the compliance of
sound particle velocity w plays the same role of a current. The
the back-chamber. Even if not present in Fig. 2, the complete
dynamics inside the sensor are described by the Newton law:
 equivalent electric model should include package impedance
dv representing the flow of the air through the inlet hole of the
F =m + bv + k vdt
dt package into the front-end cavity. L pk , R pk , and C pk account
where m is the mass, b is the friction, and k is the elastic for the air mass into the inlet hole, the viscous damping of the
modulus. In view of these analogies mass, friction, and elastic wall of the inlet hole, and the compliance of the volume of
modulus correspond to an inductance L=m, a resistance R=b, the cavity, respectively. Table I reports the estimated values of
and a capacitance C=1/k, respectively. It follows that the the components of Fig. 2 [6]. Mechanical compliance of the
transducer can be modeled by the analogous electrical circuit membrane, resistance of air gap, acoustic and ventilation holes
reported in Fig. 2. Using Fig. 3 as a reference, the physical are estimated via mechanical and/or fluidic finite elements
meaning of the components can be explained hereafter. Radi- simulations. Other values are estimated via analytical model.
ation impedance represents the interaction of the diaphragm
with the air in the front-chamber. Rr describes the frictional B. MEMS Sensor: Sensitivity and Frequency Response
force due to the radiation of sound back into the front From Fig. 2, the transfer function between input sound
chamber, whereas L r is related to the inertial mass of the pressure P and displacement of moving membrane x d
air close to the diaphragm and vibrating in phase with it. results (1), as shown at the bottom of the next page,

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 5388 IEEE SENSORS JOURNAL, VOL. 18, NO. 13, JULY 1, 2018

TABLE I
E STIMATED C OMPONENT VALUES U SED IN F IG . 2

Fig. 5. Transducer noise spectral density.

The low frequency behavior and the high-pass cutoff fre-

quency are linked to ventilation hole and back chamber
geometries:
1 ∼
fH P =   = 85H z
2π Csp + Cbc RV
Since environmental noise is typically dominated by low
frequency contributions such as wind noise, door slam, or also
tapping on the sound port, the designed cut-off frequency is
able to (partially) suppress these kind of acoustic disturbance.
The resonance frequency, quality factor and the high fre-
quency behavior are someway related to all parameters of the
model, with membrane the dominant one:
Fig. 4. Electrical circuit for transducer’s mechanical and acoustical parts. 1 ∼
fres =  = 50.9k H z
C sp C bc
2π C sp +C bc (L d + L A )

where R A = Ra + Rbp + Rr + R pk , L A = L a + L bp + L r + L pk , 
1  Ld + L A ∼
and Sd is the effective capacitive area (0.89mm2) of the Q=  C C = 22.6
RA sp bc
diaphragm which is larger than the simple difference between C sp +C bc
the surface of the back-plate and the acoustic holes because The behavior in the audio passband, i.e. the transducer
the fringing electric fields present at the border of the holes sensitivity, is given by:
reduce their effective diameter. Since most capacitive sensor
are charge biased by a charge pump circuit VB (14.2V) through Csp Cbc V B ∼
S = Sd = 12.7mVrms /Pa → −38d BV /Pa
an extremely high impedance (in the T range), transducer Csp + Cbc x o
output voltage VO is:
VO ∼ xd C. MEMS Sensor: Noise Contributions
= VB (2)
P xo To see the effects of the noise sources to the output sensor
where x o is the biased air gap (5.6μm), i.e. the physical gap voltage, noise voltage generators were added in series to
reduced by the attractive effect resulting from the electrostatic resistors R A and RV in the equivalent circuit model, and the
force across the gap when counterbalanced by the elastic output spectral noise density was obtained in Fig. 5. One can
force of the springs. Using (1) and (2), the capacitive sensor notice that noise arising from R A is subjected to the same
sensitivity and frequency response are depicted in Fig. 4. transfer function as the input sound pressure, whereas noise

xd ∼ sCsp Cbc RV
= Sd  
  
P C C C C
1 + s Csp + Cbc RV · 1 + s Cspsp+Cbcbc R A + CspC+C
bc
bc
LA
RV + s 2 Cspsp+Cbcbc (L A + L V )

∼ sCsp Cbc RV
= Sd  
 
C C

C C
 (1)
1 + s Csp + Cbc RV · 1 + s Cspsp+Cbcbc R A + s 2 Cspsp+Cbcbc (L A + L V )

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 NICOLLINI AND DEVECCHI: MEMS CAPACITIVE MICROPHONES 5389

Fig. 7. Measured microphone output leakage.

and 4th integrator feedbacks, respectively. A 1st order loop

has no stability problems and can guarantee good dynamic
Fig. 6. 4th -Order 1b  Schematic. behavior for input levels close to the reference one. Moreover,
due to the pseudo random nature of voice and audio signals,
this progressive adaption of the order does not introduce
coming from RV has a low-pass behavior. Since Rbp is the spurious tones. The total in-band  noise contribution is
dominant part of R A , it can be concluded that acoustic holes about 10dB lower than the noise coming from the transducer.
contribute to the noise in the upper part of the audio band,
whereas ventilation holes dominate in the lower and middle III. H IDDEN I SSUES IN MEMS M ICROPHONES
ones. More precisely, acoustic hole and ventilation hole noise MEMS microphone producers and publications compare
contributions are 2.3μVrms and 12.8μVrms, respectively, for a microphone performance in term of parameters such as
total noise of 13μVrms on the 20Hz-20kHz audio bandwidth. power consumption, sensitivity, SNR, AOL/THD, and start-up
To account for the relative loudness perceived by the human time [2]–[4], [7]–[10]. However, there are other characteristics,
hearing A-weighting was applied to spectral densities and then never disclosed or mentioned up to now, that can influence
integrated up to 20kHz resulting to a noise of 5.3 μVrms the attractiveness of a commercial MEMS microphone. In this
A-weighted, and a SNR of 67dBA well in line with the state of section, evidence of three “hidden” issues is reported. Two
the art of capacitive microphone transducers [2]–[4], [7]–[10]. of them, that is Power-Supply Rejection degradation at low
frequencies and contamination from Radio Frequency Inter-
ference, are common to both analog and digital microphones.
D. Readout Circuit
The third one, called “swirling” noise, is present only in
Nowadays, microphones have increased their capability to digital microphone when working in stereo configuration.
capture pressure levels up to AOL=130dBSPL (36dBPa) with Even if these issues are more or less evident in all commercial
distortion down to 4% or lower. Since −38dBV/Pa corre- products, the following measurements are taken from the very
sponds to about 12.7mVrms/Pa, an AOL=36dBPa results in first realizations of analog and digital MEMS microphones of
nearly 0.8Vrms or 1.13Vp. Since a high-input impedance and our company.
large output current capability are also mandatory, to cope
with these requirements a three-stage Nested-Miller Differen- A. Power-Supply Rejection (PSR)
tial Difference Amplifier with Dynamic Resistive Degenera- Common data sheet specs require PSR > 60dB in the
tion (DDA-DRD) has been used [11]. This technique resulted audio band, resulting to an output leakage of −86dBV with
into a THD lower than 1.1% up to 130dBSPL, without a 140mVpp sinewave injected on the power-supply. Even if
introducing noise at small input levels. The idle-channel noise, simulated and measured values of the readout circuit alone,
reported as white dotted line in Fig. 5, is about 1.9μVrms charge-pump included, were much lower than this limit,
A-weighted, well below the transducer noise. Fig. 7 reports that the measured analog MEMS microphone
To summarize, the SNR performance is dominated by the leakage (blue curve) degrades at low frequencies till to reach
sensor noise, whereas AOL and THD behaviors depend on the about −76dB at 20Hz. A similar behavior is present in a
preamplifier only. In case of digital microphones, 4th order digital one. Since these measurements have been done in
1b  topologies are usually chosen to keep quantiza- an anechoic chamber [12], it has been noticed that this
tion noise well below sensor noise in the audio bandwidth, degradation gets smaller by reducing the pressure inside
and to save area and consumption with respect to multibit the chamber, till to disappear when the pressure is around
 modulators followed by 5th order digital noise shapers. 850μBar (orange curve). Moreover, a sudden switch between
In our solution a cascade of four integrators with weighted blue and orange plots happens when the charge-pump circuit
feedforward summation (CIFF) topology has been selected, that biases the membrane is powered down, i.e. when pressure
as shown in Fig. 6 (do not consider for now the blue box). to voltage transduction is stopped.
Even if not shown, each amplifier is equipped with a
circuit that senses when its output signal crosses a threshold B. Radio Frequency Immunity (RFI)
and progressively reduces the order of the loop filter from RFI test requires putting in proximity of the microphone
4-th to 1-st by inserting SC dumping on the 2nd , 3rd , package a strip-line with a 217 Hz modulated RF signal. This

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 5390 IEEE SENSORS JOURNAL, VOL. 18, NO. 13, JULY 1, 2018

Fig. 10. Stereo configuration: swirling noise spectrogram on both channels.

Fig. 8. Measured RFI vs carrier frequency.

Fig. 11. Thermal equivalent circuit.

Fig. 9. Stereo configuration: swirling noise. As for the PSR issue, it has been noticed that the bump
in the frequency spectrum and the spectrogram trace due
to swirling noise get smaller and smaller by reducing the
test emulates the electromagnetic coupling from cellular phone pressure inside the anechoic chamber, till to disappear when
TX antenna and a possible demodulation of the TDMA pattern the pressure is around 850μBar.
by a non-linear behavior inside the MEMS microphone. Even
if RFI specs are still under debate today (in most data sheet
IV. O RIGIN OF THE H IDDEN I SSUES
they are missing), one can expect that demodulated power
should be comparable to or lower than microphone noise This section explains causes of issues reported in Section III.
integrated in the audio bandwidth to avoid TDMA buzz noise
annoyance. Considering that the attenuation of the A-weighted
A. PSR Worsening at Low Frequency
filter is more than 10dB at 217Hz, a demodulated signal
of −100dBV results to about 3μVrms , comparable to the Basically, modulation of the power supply voltage by a
noise floor of high performance commercial MEMS micro- sinewave results in a modulation of the readout circuit power
phones [2]–[4]. As a consequence, in our opinion, −100dBV dissipation and in turn of the temperature inside the package
should be considered a reasonable upper limit for RFI spec. cavity (back-chamber) of the MEMS microphone. Since the
Fig. 8 depicts the level of the demodulated 217Hz at the analog back-chamber volume is fixed, according to the ideal gas
microphone output (blue curve) versus RF carrier frequency. law the temperature variation directly translates to pressure
A similar behavior can be found in a digital microphone. change, which in turn is collected and transduced by the
Even if it was not possible to replicate the RFI test in a MEMS sensor. This can be better understood thanks to the
chamber with reduced pressure, however it has been noticed very basic thermal model of Fig. 11, where P is the power
that the level of the demodulated signal strongly reduces by dissipated by the readout circuit, Cth and Rth are, respectively,
powering down the charge pump circuit (red curve). the overall thermal capacitance and resistance accounting for
the temperature changes inside the back-chamber.
C. Swirling Noise (SN) Since extraction of the thermal parameters by package
simulation is too complex and thus not completely reliable,
In a first realization of our digital microphone, when the Cth and Rth values have been determined empirically, as
channel is idle you can hear (after external decimation and D/A described later on. When a sinewave is superimposed on
conversion) a welcome or comforting noise if the microphone the power supply it results to a modulation of the power
is in mono configuration. However, in stereo configuration, dissipation Qd =Icc Vrms , where Icc is the readout circuit
on few samples there is evidence of an annoying disturbance static current and Vrms is the rms value of the sinewave signal,
which resembles swirling noise. These tones can be seen both and in turn a temperature change T inside the back-chamber
in the frequency spectrum, as shown in Fig. 9, and in the equal to:
spectrogram performed on the bit-stream of the two channels,
as shown in Fig. 10. Even if it is hard to distinguish them Rt h
Tbc = Q d (3)
from noise floor, they can easily be heard. 1 + s Rt h C t h

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 NICOLLINI AND DEVECCHI: MEMS CAPACITIVE MICROPHONES 5391

formulae (3) to (5), for example by estimating the increment

of Vout_leak when Icc is trimmed from 180μA to 340μA, as
reported hereafter:
300mV pp
Q d = Vrms · ICC = √ · (340μA − 180μA)
2 2
= 17.02μWrms
Rt h ∼ 1
Tbc = Q d = Q d
1 + s Rt h C t h sCt h
1
= 17.02μWrms · = 27.4μ◦ K
2π · 33H z · 3m J/◦ K
P 100k Pa
Pbc = Tbc = · 27.4μ◦ K = 9.12m Pa
Fig. 12. Measured microphone output leakage. T 300K ◦
Vout _leak = Pbc · S( f = 33H z)
∼ 33
= 9.12m Pa · (12.6mVrms /Pa) · = 44.6μVrms
85
This value is in close agreement with the measured difference
85μVrms − 40μVrms =45μVrms reported in Fig. 13.

B. TDMA Burst Demodulation

Two different mechanisms have been found responsible of
the RFI problem in MEMS microphones, i.e. conducted and
radiated electromagnetic interference (EMI).
Conducted interference: when a GSM carrier “enters” the
microphone package through its pins, for example the two
Fig. 13. Measured microphone output leakage vs readout static current.
outputs and the power supply in an analog MEMS microphone,
non-linear behaviors in these paths lead to demodulation of
the TDMA burst. This in turn results to a dominant thermal
From the ideal gas law PV=NkB T, where P is the pressure, coupling and secondary electrical coupling in the microphone
V the volume, T the temperature, kB the Boltzmann con- channel. Thermal coupling arises from the variation of the
stant, and N the number of atoms/ molecules in the gas, it power dissipation of the readout circuit. This, as explained in
follows that: Section IV.A, leads to a pressure variation in the back-chamber
which directly translates to an electrical signal at the MEMS
P
Pbc = Tbc with P = 100kPa and T = 300◦ K (4) microphone output.
T Radiated interference: electromagnetic power associated to
Since, as shown in Fig. 2, a pressure change inside the back- the GSM carrier is also dissipated inside the microphone
chamber has the same transfer function of a pressure signal package, heating the back-chamber air. However, since the
at the microphone input, the output signal that leaks from the thermal behavior of the package is governed by very high
power supply is given by: time constants, as explained in Section IV.A, this power is
P Rt h spectrally focused at the envelope frequency, i.e. the modula-
Vout _leak = Q d · S( f ) (5) tion signal frequency, resulting to an extra thermal coupling
T 1 + s Rt h C t h
in the microphone input. Both conducted and radiated thermal
As a consequence, the transfer function from the power
couplings are evident in Fig. 8 of Section III.B where the blue
supply to the microphone has a bandpass behavior. It starts
curve moves to the red one when the charge pump is powered
with a +20dB/decade slope at very low frequencies, then
down, zeroing the sensor sensitivity. Indeed, the red curve is
there is the “thermal” pole pth =1/Rth Cth and finally the pole
evidence of the electrical coupling from the conducted signal
related to sensitivity around 85Hz. This is experimentally
inside the readout circuit.
verified in Fig. 12 that reports the measured microphone output
leakage, with frequency range expanded to 0.1Hz, when a
300mVpp sinewave is superimposed to the power supply. C. Swirling Noise in Stereo Configuration
Looking at the plot, Cth is derived from the measured data, for In the stereo mode configuration the L/R select pin allows
instance, at 300Hz, whereas Rth from the first pole once Cth is to multiplex two different microphone bit streams on a single
known. data output wire, as shown in Fig. 14a. Assuming that L/R
To furtherly confirm that the bad PSR at low frequency select pins of the left (L) and right (R) microphones are set to
is a thermal related effect, Fig. 13 shows that the measured GND and VCC, respectively, the L channel transmits data just
output leakage from a 33Hz 300mVpp sinewave injected on the after the falling clock edge and is in high impedance (HI) just
power supply is almost linearly related to the static current before the following rising clock edge, whereas the R channel
of the readout circuit. This result can be well predicted by is active just after the rising clock edge and is HI just before

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 5392 IEEE SENSORS JOURNAL, VOL. 18, NO. 13, JULY 1, 2018

Fig. 16. Measured microphone output leakage.

Fig. 14. Microphone stereo mode operation.

Fig. 17. EMI filters.

Fig. 18. Measured RFI vs carrier frequency with EMI filters.

Fig. 15. Simulated swirling noise a) audio band FFT and b) spectrogram.

channels obtained after the XOR of the two bit streams

the following falling clock edge. Even if optimum dithering followed by applications of formulae (3)-(5).
techniques [13] have been used to remove in-band idle-channel
patterns in  modulators with minimum SNR penalties, V. P OSSIBLE S OLUTIONS AND E XPERIMENTAL R ESULTS
it may happen that some idle-channel tones are still present on
the output bit streams around half of the sampling frequency. A. Glob Top
Since parasitic capacitance CL loading L/R data output can be Since low frequency PSR is linked to a thermal effect, it can
in the range of hundreds of pF, large output currents IOUT_R be improved if the overall thermal constant Cth Rth is increased
and IOUT_L (dis)charges CL when two adjacent bits, belonging by covering the readout circuit ASIC with glob top. Glob top
to L and R channels, have different binary values, whereas no is a high viscosity epoxy or silicone material normally used
current is needed when two adjacent bits have the same binary in non-hermetic packaged ICs for environmental protection,
value, as depicted in Fig. 14b. such as corrosion, moisture, micro-contamination, light, etc.
Basically, IOUT_R and IOUT_L resemble a XOR operator Glob top effect is two sided. First, it adds thermal mass
between the left and right bit streams and, since this relation- to the ASIC, increasing its thermal capacitance, and then it
ship is intrinsically non- linear, the high frequency tones are increases thermal resistance from silicon to air. Both factors
folded in the audio band by intermodulation and appear in the reduce the magnitude of thermal changes inside the package.
left- and right-channel output current spectrum. This leads to Moreover, since glop top size is only a small percentage of the
modulation of the power supply dissipation Qd =VCC IOUT , back-chamber, the reduction of sensor sensitivity is negligible.
resulting to back chamber temperature and pressure changes A comparison between standard (blue curve) and glob top
according to (3) and (4), respectively, and pollution of the (red curve) analog microphone is shown in Fig. 16. Since
microphone signal according to (5). glob top achieves more than 10dB reduction of the output
Fig. 15 reports the simulated in-band spectra of the L/R microphone leakage from power supply disturbances at very
channels and the swirling noise spectrogram on one of the low audio frequencies (around 20Hz), one can expect to extend

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 NICOLLINI AND DEVECCHI: MEMS CAPACITIVE MICROPHONES 5393

TABLE II
P ERFORMANCE C OMPARISON W ITH S TATE - OF - THE A RT

Fig. 19. Stereo configuration: spectrogram with FIR DAC.

Fig. 20. Analog microphone prototype photograph.
this behavior even at higher frequencies (dotted red line) until
crossing the flat portion of the curve where the ‘electrical’ PSR
dominates on the ‘thermal’ one. However, in our case it was
not possible to protect the ASIC, for instance covering with a
shielding layer star-routed to GND pad, because the last metal
was already used for power supplies. For this reason, glob
top created unwanted coupling paths between noisy power
supply or charge-pump nodes and the high impedance pads
connected to the sensor which seriously impacted the PSR in
the whole audio band (Fig. 16, red curve).
Clearly, this limitation can be avoided by a complete re- Fig. 21. Digital microphone prototype photograph.
layout of the readout circuit in future versions to get full
benefit from glob top option in mass production. eliminate these tones, with minor penalty of in-band SNR,
is to put a FIR DAC in the  modulator feedback to realize
B. EMI Filters a notch filter at half of the sampling frequency, as shown in
The conducted interference can be strongly reduced by the blue box of Fig.6. Since the FIR DAC is made by one
inserting passive filters on the two outputs and the power flip-flop only and its coefficients k1 and k2 (with k1 +k2 =1)
supply in an analog MEMS microphone (a similar solution are simply realized by splitting the VREF sampling capacitors
has been adopted for the pins of the digital microphone). accordingly, extra area and consumption is negligible. Thanks
Each filter, made by a resistor and a capacitor integrated on to this solution no swirling noise was heard anymore, as also
the common substrate, is implemented between a package depicted in Fig. 19 that reports the spectrogram performed on
pin and its related bond-finger, as shown in Fig. 17, in the bit-streams of the two channels.
order to attenuate the coupled RF carrier prior to TDMA
burst demodulation inside the ASIC. It follows that both D. Summary of the Measured Results
thermal and electric couplings linked to the conducted inter-
Both analog and digital microphone prototypes have been
ference are strongly reduced, as shown in Fig. 18. The green
implemented, as shown in Fig. 20 and 21, respectively.
curve shows that residual thermal coupling, probably due to
In the case of analog microphone the sensor is made by
the radiated interference, still dominates on the electric one
two parallel 0.5mm × 1.0 mm polysilicon membranes with
(blue curve).
effective air gap of 5.6μm and serpentine spring structures
placed at the corners. For the digital microphone, membrane
C. FIR DAC area and air gap are scaled approximately by a factor of
Since origin of swirling noise is linked to idle-channel tones two. The readout ASICs are realized in a 0.5μm single-
around half of the sampling frequency, an effective way to poly, four metal layers, N-well CMOS process with MIM

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.
 5394 IEEE SENSORS JOURNAL, VOL. 18, NO. 13, JULY 1, 2018

capacitors option. The analog ASIC die area is 0.98mm2, [6] A. Faes, “Micromachined condenser microphone with piston
current consumption is about 220μA with a power supply diaphragm,” Ph.D. dissertation, ITC-IRST, Trento, Italy, 2003.
[7] J. Citakovic et al., “A Compact CMOS MEMS Microphone with 66 dB
ranging from 2.4V to 3.6V. The analog microphone sensitivity SNR,” in IEEE ISSCC Dig. Tech. Paper, Feb. 2009, pp. 350–351.
is about −38dBV/Pa, the SNR is nearly 67dBA at 1Pa, and [8] S. Ersoy, R. H. M. van Veldhoven, F. Sebastiano, K. Reimann, and
the AOL is more than 130dBSPL (36dBPa) at power supplies K. A. A. Makinwa, “A 0.25 mm2 AC-biased MEMS microphone
interface with 58 dBA SNR,” in IEEE ISSCC Dig. Tech. Papers,
higher than 2.75V. The digital ASIC die area is 1.2mm2, Feb. 2013, pp. 382–383.
current consumption is about 720μA with a power supply [9] S. A. Jawed, D. Cattin, M. Gottardi, N. Massari, A. Baschirotto, and
ranging from 1.6V to 3.6V. The digital microphone Sensitivity A. Simoni, “A 828 μW 1.8 V 80 dB dynamic-range readout interface for
a MEMS capacitive microphone,” in Proc. ESSCIRC, Edinburgh, U.K.,
is about −26dBFS, the SNR is nearly 63dBA at 1Pa, and Nov. 2008, pp. 442–445.
the AOL is more than 120dBSPL (26dBPa) at power supplies [10] AD Inc. (2012). ADMP401 Omnidirectional Microphone with Bot-
higher than 1.8V. Measurements of other parameters which tom Port and Analog Output—Data Sheet. [Online]. Available:
www.analog.com
anyway are of top importance for the acoustic behavior of the [11] A. Barbieri and S. Pernici, “A differential difference amplifier with
microphone such as PSR and RFI are reported in part A and B dynamic resistive degeneration for MEMS microphones,” in Proc.
of this Section. Table II reports performance summary and ESSCIRC, Losanne, Switzerland, Sep. 2016, pp. 285–288.
[12] Acoustic System Device SD2 Test Enclosure-Data Sheet, RF-2300,
comparison with state-of-the art. ETS Lindgren, Inc., 2017. [Online]. Available: www.ets-lindgren.com
[13] S. R. Norsworthy, “Effective dithering of sigma-delta modulators,” in
Proc. ISCAS, San Diego, CA, USA, May 1992, pp. 1304–1307.
VI. C ONCLUSION
This paper has presented three ‘hidden’ issues linked to
the assembly of the MEMS sensor and its readout circuit in
the same package. Power-Supply Rejection degradation at low Germano Nicollini was born in Piacenza, Italy.
He received the Laurea degree in electronic engi-
frequencies and pollution from Radio Frequency Interference neering from the Universita’ di Pavia, Pavia, Italy,
are common to both analog and digital microphones, whereas in 1981. From 1982 to 2008, he was with STMi-
“swirling” noise is present only in digital microphones when croelectronics, Italy, where he was involved in
and managed the design of analog and mixed-
working in stereo configuration. Even if all commercial prod- signal ICs for wired and wireless telecommuni-
ucts present evidence of these issues, they have never been cation. From 1990 to 1992, he was a Visiting
reported in data sheet or discussed up to now. To quantify Professor with the Universita’ di Genova, Genova,
Italy, and was with the ISSCC Technical Pro-
performance degradations, causes, and possible solutions, we gram Committee from 2006 to 2008. From 2008 to
have used the very first realizations of analog and digital 2013, he was with ST-Ericsson, Italy, as a Technical Director of the Audio
MEMS microphones of our company. and Converter Expertise Center. In 2013, he joined the Audio, MEMS, and
Sensor Group at STMicroelectronics, Cornaredo, Italy, where he is currently
a Company Fellow. He has been granted 65 patents and has several pending
ACKNOWLEDGMENT patents. He has specialized in the areas of sigma-delta conversion for audio
and RF baseband processing, and acoustic interfaces, and is a coauthor of
The authors would like to thank S. Adorno, S. Conti, several conference and journal papers in these fields.
S. Pernici, A. Barbieri, E. Marino, F. Barbieri, A. Danioni,
E. Borlini, F. Guerinoni, and L. Gasparetto for the their
valuable help.

R EFERENCES Daniele Devecchi was born in Milano, Italy.

He received the Laurea degree in electronic engi-
[1] J. J. Neurmann and K. J. Gabriel, “A fully-integrated CMOS-MEMS neering from the Politecnico di Milano, Milano,
audio microphone,” in Proc. Solid-State Sens., Actuators, Microsyst., in 1979. From 1979 to 2008, he was with STMi-
2003, pp. 230–233. croelectronics, Italy, where he was involved in and
[2] AKU440 Bottom Port, Akustica, Inc. (2014). Digital Silicon MEMS managed the design of analog and mixed-signal
Microphone—Data Sheet. [Online]. Available: www.akustica.com ICs for consumer and ADSL and wired telecom-
[3] SPH0641LM4H-1 SiSonic MEMS Microphones, LLC. (2014). Data munications. From 2008 to 2013, he was with ST-
Sheet Knowles Electronics. [Online]. Available: www.knowles.com Ericsson, Italy, where he was responsible for the
[4] E. Bach et al., “A 1.8 V true-differential 140 dB SPL full-scale standard validation laboratory. In 2013, he joined the Audio,
CMOS MEMS digital microphone exhibiting 67 dB SNR,” in IEEE MEMS, and Sensor Group at STMicroelectronics,
ISSCC Dig. Tech. Papers, Feb. 2017, pp. 166–167. Cornaredo, Italy, where he is currently a senior expert in the development of
[5] K. Heutschi, Electrical-Mechanical-Acoustical Analogies (Lecture Notes ASICs for MEMS microphones, gyroscopes, and micro-mirrors. He has been
on Acoustic II). ETH Zurich, Switzerland, Jan. 2017. granted 6 patent and is co-authored five publications.

Authorized licensed use limited to: Vel Tech Dr. RR & Dr. SR Tech University - Chennai. Downloaded on August 27,2020 at 09:25:49 UTC from IEEE Xplore. Restrictions apply.