1

The UmboMic: A PVDF Cantilever Microphone

Aaron J. Yeiser*, Emma F. Wawrzynek*, John Z. Zhang, Lukas Graf, Christopher I. McHugh, Ioannis Kymissis,
Elizabeth S. Olson, Jeffrey H. Lang*, Hideko Heidi Nakajima*

Abstract—Objective: We present the “UmboMic,” a prototype

piezoelectric cantilever microphone designed for future use with
totally-implantable cochlear implants. Methods: The UmboMic Umbo
sensor is made from polyvinylidene difluoride (PVDF) because
of its low Young’s modulus and biocompatibility. The sensor
is designed to fit in the middle ear and measure the motion Incus
arXiv:2312.14339v1 [eess.AS] 22 Dec 2023

of the underside of the eardrum at the umbo. To maximize

its performance, we developed a low noise charge amplifier
in tandem with the UmboMic sensor. This paper presents the
performance of the UmboMic sensor and amplifier in fresh
cadaveric human temporal bones. Results: When tested in human
temporal bones, the UmboMic apparatus achieves an equivalent
input noise of 32.3 dB SPL over the frequency range 100 Hz
to 7 kHz, good linearity, and a flat frequency response to Stapes Triangular
within 10 dB from about 100 Hz to 6 kHz. Conclusion: These Sensor Tip
results demonstrate the feasibility of a PVDF-based microphone
when paired with a low-noise amplifier. The reported UmboMic
apparatus is comparable in performance to a conventional
Fig. 1: An image of a human cadaveric middle ear cavity with
hearing aid microphone. Significance: The proof-of-concept the UmboMic sensor inserted. The UmboMic sensor tip is
UmboMic apparatus is a promising step towards creating touching the underside of the eardrum at the umbo. The umbo
a totally-implantable cochlear implant. A completely internal is the point where the end of the manubrium of the malleus
system would enhance the quality of life of cochlear implant attaches to the tympanic membrane. The malleus, incus, and
users.
stapes make up the ossicular chain.

I. I NTRODUCTION
OCHLEAR implants are a successful neuroprosthetic sound localization cues derived from the outer ear structure.
C that can restore hearing to people with severe
sensorineural hearing loss. While partially implanted, they rely
Engineering a practical internal microphone would enable a
totally-implantable cochlear implant. Although development
on an external hearing aid microphone that is positioned on of implantable microphones has been ongoing for years,
the side of the head. The external nature of this microphone none are currently on the market. Technical approaches range
imposes many lifestyle restrictions on cochlear implant users. from fiber-optic vibrometery [1] to capacitive displacement
Patients cannot swim or play certain sports while wearing the sensing [2]. Two devices are currently in clinical trials: a
external unit, nor can they wear it while sleeping. Additionally, piezoelectric sensor called the Acclaim by Envoy [3] [4] and a
an external microphone does not provide the pressure gain and subcutaneous microphone called Mi2000 by MED-EL [5] [6].
There is very little information available about either device,
Submitted December 2023. This paper is partially supported by NIH Grant and they remain in testing.
R01 DC016874, NSF GRFP Grant 1745302, NSF GRFP Grant 2141064, a
grant from the Cloëtta Foundation, Zürich Switzerland, and the Research Fund The microphone reported here is a piezoelectric sensor
of the University of Basel, Switzerland. paired with a charge amplifier that we call the “UmboMic”.
A. J. Yeiser was with the MIT Electrical Engineering and Computer Science We refer to the piezoelectric sensor as the “UmboMic sensor”
(EECS) Department, Cambridge, MA, 02139. He is now an independent
engineering consultant (e-mail: ayeiser@mit.edu). and the sensor connected to the amplifier as the “UmboMic
E. F. Wawrzynek is with the MIT Electrical Engineering and apparatus.” The UmboMic sensor detects the motion of the
Computer Science (EECS) Department, Cambridge, MA, 02139. (e-mail: umbo, which is the tip of the malleus that attaches to the
emmafw@mit.edu).
J. Z. Zhang is with the MIT Mechanical Engineering Department (e-mail: conical point on the underside of the eardrum. Figure 1 shows
johnz@mit.edu). a picture of the UmboMic sensor in contact with a human
L. Graf is with the Harvard Medical School and Mass. Eye and Ear umbo. Umbo displacement is large for all auditory frequencies
Department of OHNS (e-mail: lukas.graf@usb.ch)
C. I. McHugh is with the Harvard Medical School and Mass. Eye and Ear and mostly unidirectional in humans, making it an ideal target
Department of OHNS (e-mail: christopher mchugh@meei.harvard.edu) for sensing motion. By sensing the umbo, the UmboMic
I. Kymissis is with Columbia University Department of Electrical apparatus has an advantage over microphones that target other
Engineering (e-mail: johnkym@ee.columbia.edu)
E. S. Olson is with Columbia University Departments of OTO/HNS and parts of the ossicular chain. For example, the Acclaim by
Biomedical Engineering (e-mail: eao2004@cumc.columbia.edu) Envoy targets the incus body, which has complex modes of
J. H. Lang is with the MIT EECS department (e-mail: lang@mit.edu). motion around 2 kHz.
H. H. Nakajima is with the Harvard Medical School and Eaton-
Peabody Laboratories at Mass. Eye and Ear Department of OHNS (e-mail: We build the UmboMic sensor out of a thin film
heidi nakajima@meei.harvard.edu). piezoelectric polymer called polyvinylidene difluoride
 2

Conductive ink Piezoelectric layer

50 um PVDF
Ground Electrode
Adhesive
PCB pad Ground Shield 5-10 um epoxy
(connected to gound shield)
(100-200 nm sputter coated metal)

Ground Electrode Flex PCB substrate

PCB pad 100 um polyimide
(connected to gound shield) (Border around electrodes
not shown)

Positive Electrode PCB pads are mirrored on the

Positive Electrode
opposite side of the board
PCB pad 100-200 nm sputter coated metal
(with solder mask to prevent Negative Electrode
short to ground shield)
100-200 nm sputter coated metal

Fig. 2: Differential PVDF cantilever diagram (not to scale). We construct the cantilever from two layers of PVDF sandwiching
a flex PCB base. We sputter-coat the charge sense electrodes onto the flex PCB and capacitively couple the electrodes to the
PVDF through the thin glue layer. Note that the border around the electrodes is not shown in this figure.

(PVDF). PVDF is excellent for our application because it the sensor tip. The triangular shape is a design commonly used
is highly flexible and biocompatible [7]. Typically, PVDF with piezoelectric sensors and actuators [10] as it increases
is considered a poor choice for small-area sensors as it is the sensor’s robustness by equalizing stress concentration. A
less sensitive than piezoelectric ceramics. To overcome this triangular shaped sensor is also practical given the anatomical
limitation, our design relies on the differential measurement limitations of the middle ear. The sensor’s tapered shape allows
between two layers of PVDF connected to an extremely low it to slide into position without hitting the other ossicles during
noise amplifier to boost the signal-to-noise ratio. This paper insertion.
presents a prototype PVDF sensor and an accompanying There are a few factors to consider when deciding on
custom low-noise differential charge amplifier. The reported UmboMic sensor geometry. Our sensors must be small enough
UmboMic apparatus exhibits high sensitivity and low to fit through a variety of middle-ear cavity surgical entrances.
noise comparable to commercially available hearing-aid However, in order to maximize the charge output of our
microphones such as the Sonion 65GG31T [8] and the piezoelectric sensor, we want its active surface area to be as
Knowles EK3103 [9]. In the next steps, we are advancing large as possible. A larger sensor is also faster and cheaper
the microphone with fully biocompatible, decades-durable to fabricate. We found through testing that a 3 mm by 3 mm
materials. triangular sensor tip fits well within the middle ear cavity of
multiple cadaveric specimens, and is large enough to produce
II. C ANTILEVER DESIGN AND FABRICATION a sufficient output charge.
The UmboMic sensor is a triangular bimorph cantilever Further details on the UmboMic’s sensor design are detailed
approximately 3 mm wide at the base, 3 mm long, and 200 µm in [11].
thick. The free end of the triangular tip interfaces with the
umbo to sense its motion. We design the UmboMic sensor B. Electrode patterning
to have a relatively uniform stress distribution in the PVDF. To simplify the fabrication of the UmboMic sensor, we use
The UmboMic sensor is fabricated with two layers of 50 µm a flex PCB as the base substrate of the sensor. The custom
PVDF sandwiching a 100 µm Kapton flexible printed-circuit- flex PCB has a polyimide core with electrode and ground
board (flex PCB) substrate; this construction is detailed in traces connecting to a U.FL connector solder footprint. We use
the following sections and in Figure 2. The use of a Kapton photolithography to pattern triangular charge sense electrodes
flex PCB as the core layer greatly simplifies attaching cables at the top of the flex PCB substrate, and this constitutes the
to the device. Additionally, the PCB design allows for the active region of our sensor.
ground electrode to double as a ground shield, which works in Through experimentation we found that cantilever designs
tandem with the differential sensor output to nearly eliminate with charge sense electrodes exposed to the outside of the
electromagnetic interference. sensor tend to have unacceptably high parasitic leakage
conductance, especially in wet environments like the middle
A. Designing sensor dimensions ear cavity. Our fabrication strategy revolves around pre-
We use a triangular shape for the UmboMic sensor as it patterning the charge sense electrodes and then trimming the
results in a uniform stress and charge distribution throughout sensor to leave a margin around the electrodes, eliminating
 3

(a) Left to right: We sputter-coat 200 nm of aluminum onto a PCB

and then pattern into the shape of the electrodes using contact
photolithography.

Fig. 4: The assembled differential amplifier board.

the stages of electrode deposition and patterning.

C. PVDF adhesion
(b) We apply conductive ink at the interface between the patterned Before gluing the PVDF film to the sputter coated metal,
electrode and the PCB electrode pattern. We then use Kapton tape we reinforce the electrical connection between the patterned
to mask off areas of the flex PCB and glue the PVDF on each side
electrodes and the flex PCB traces with a silver conductive
with epoxy. The PVDF is the transparent film seen in the image.
ink pen. We then sand one side of the PVDF with 3000-
grit sandpaper to increase surface roughness and mask the
portions of the flex PCB that must remain glue-free. Next, we
generously apply epoxy between the two PVDF layers and
the flex PCBs. Devcon Plastic Steel epoxy works well for
bonding the PVDF to the polyimide substrate. We orient the
piezoelectric films such that they have opposing polarization.
Finally, we squeeze as much epoxy as possible out from
between the flex PCB and PVDF film with a doctor blade,
(c) We trim the sensor to its final shape. Note the small border around
the patterned electrode and the stackup is left to cure. This method achieves a 10 µm
epoxy thickness, which is sufficiently thin to allow efficient
capacitive coupling. The masking and bonding process is
shown in Figure 3b.

D. Finishing steps
After the epoxy is cured, we trim the PVDF and flex PCB to
shape with scissors leaving a buffer of approximately 300 µm
between the edge of the electrode and the edge of the PCB
(d) We sputter coat a second layer of 200 nm aluminum onto the layer of the sensor, shown in Figure 3c. This buffer serves
sensor, forming the ground shield. to protect the electrodes from water ingress, which could
otherwise short the sensor. We then encapsulate the sensor
Fig. 3: The fabrication process for the UmboMic sensor.
tip with a 200 nm layer of sputter-coated aluminum. This
outer layer serves as both a ground electrode and a ground
shield, protecting the charge sense electrodes from EMI. We
this leakage path and improving the UmboMic apparatus noise connect the step between the PVDF and the flex PCB with
floor. conductive ink or adhesive as shown in Figure 3d. This ensures
We first apply 200 nm of aluminum to both sides of the flex the aluminum on the PVDF layer is electrically connected to
PCB using an AJA sputter coater. Next, we spin-coat a layer the ground pad on the PCB at the tail. Finally, we solder a
of AZ3312 positive photoresist on both sides of the sputter- U.FL receptacle on either side of the tail end of the UmboMic
coated PCB, bake for approximately two minutes at 110 ◦ C, sensor opposite from the electrodes. Figure 2 shows the stack-
place in a contact photolithography mask, and flood expose up of the tip of the finished UmboMic sensor.
for 30 seconds on each side. We then dissolve the UV-exposed
photoresist and the aluminum underneath using a tetramethyl III. D IFFERENTIAL C HARGE AMPLIFIER
ammonium hydroxide (TMAH) solution. Finally, we dissolve It is imperative to device performance to achieve signal
the remaining photoresist in acetone. Figure 3a summarizes amplification without introducing too much noise. By
 4

Rf

Cgnd
PVDF Cf Ro
iin+
vin+
−
LTC6241 vout
vout−
Cpiezo + 10 kΩ 10 kΩ
(oa1)
Rpar Cpar qin iin Co
−
+ (oa1)
vpiezo + AD8617 vint
+
− vout+
LTC6241 + (oa2) AD8617
vin−
− 10 kΩ − (oa3)
iin−
Cf 10 kΩ
Cgnd Cb
Rb Ra

Rf

Fig. 5: The differential sensor (outlined in red) and charge amplifier topology. We model the piezoelectric sensor as the voltage
source vpiezo in series with the capacitor Cpiezo , together with a parasitic capacitor Cpar , leakage resistor Rpar , and capacitor
to ground Cgnd . Estimates of the piezoelecric and parasitic component values are given in Table I. Our implementation of the
differential amplifier uses Rf = 10 GΩ, Cf = 1 pF, Ra = 90 kΩ, Rb = 10 kΩ, Cb = 100 nF, Ro = 100 kΩ, and Co = 100 nF.

Cpiezo Cpar Rpar Cgnd

10 pF ∼1 pF ∼ 1 TΩ 0.6 fF
AD8617 (dual 8613) op-amp (oa2), followed by a lead gain
stage based on the AD8617 op-amp (oa3) with an output
TABLE I: Estimates of piezoelectric and parasitic component high-pass filter. The LTC6241 is chosen for its excellent
values. noise performance. The AD8617 is chosen for its good noise
performance, low bias current, and rail-to-rail operation.
The amplifier input can be interpreted as either qin , iin or
developing our own differential charge amplifier, shown in vpiezo , which are related by
Figures 4 and 5, we minimize the noise floor while providing
a gain of 20 V/pC over a −3 dB bandwidth of 160 Hz to qin = iin /jω = Cpiezo vpiezo . (1)
50 kHz. Figure 5 illustrates the charge amplifier connected to
our differential sensor having capacitance Cpiezo and charge The internal high-pass charge-to-voltage gain Gint is then
output qin . We also show parasitics Cpar , Rpar , and Cgnd : given by
parasitic parallel capacitor, leakage resistor, and capacitor
2jωRf
to ground, respectively. Estimates of the piezoelectric and vint /qin ≡ Gint = , (2)
parasitic component values are given in Table I. The charge-to- 1 + jωRf Cf
voltage gain of such a charge amplifier is invariant to parasitic which is independent of parasitics; note that the internal
resistance and capacitance, giving it good gain uniformity from current-to-voltage gain is Gint /jω. The overall gain of the
sensor to sensor. The amplifier’s differential input interfaces amplifier is given by
with our differential-mode sensor to reduce EMI. Similar
differential charge amplifiers are frequently used as low- vout /qin ≡ Gout
noise preamplifiers for high-impedance AC sources such as (1 + jω(Ra + Rb )Cb ) jωRo Co
= Gint . (3)
piezoelectric sensors [12] and charged particle counters [13], (1 + jωRb Cb ) (1 + jωRo Co )
[14].
For the component values in Figure 5, the highest high-pass
cut-on frequency 1/Rb Cb is set at 1000 rad/s (160 Hz) to filter
A. Gain Analysis out low-frequency body noise. The high-end low-pass cut-off
Our differential amplifier comprises two parallel low- frequency is set well above the audio range by the op-amp
impedance input stages based on the LTC6241 (dual LTC6240) dynamics, and so is not modeled here. Finally, the mid-band
op-amp (oa1), followed by a difference stage based on the gain is given by Gout = 20/Cf = 20 V/pC.
 5

B. Noise analysis Finally, the input-referred current variance density resulting

There are five significant noise sources in the amplifier: from the difference stage may be expressed as
Johnson noise from Rf and Rpar , voltage noise and current 4ω 2 v 2 oa2

1

2 2 2 2
noise from oa1, and voltage noise from oa2. Johnson noise i in,v,oa2 = = v oa2 + ω Cf . (13)
|Gint |2 Rf2
is treated here as a parallel current source. Being in parallel
with the input current, Rpar contributes an input-referred The total input-referred current variance density i2 in is
current variance density of 4kB T /Rpar . Together, the two Rf obtained by summing (4), (10), (12) and (13). Dividing i2 in
contribute the same input-referred current variance density by ω 2 gives the input-referred charge variance density q 2 in .
as would a single 2Rf in parallel with the input current, Finally, expanding Zdiff and Zf , and collecting terms, gives
or 4kB T /2Rf . Thus, the total input-referred current variance
2 1
density associated with Johnson noise is q 2 in = 2v 2 oa1 Ctot + q 2 oa1 + v 2 oa2 Cf2
2
4kB T 1 4kB T ι2 oa1
i2 in,johnson = . (4) + 2 + +
(2Rf ) ∥ Rpar ω (2Rf ) ∥ Rpar 2
The noise voltages at vin+ and vin− effectively produce !
2v 2 oa1 v 2 oa2
an input-referred noise current iin,v,oa1 determined by the + + , (14)
impedances of the sensor and the oa1 feedback network. While ((2Rf ) ∥ Rpar )2 Rf2
the noise voltages are not completely frequency independent, where
flicker noise for the LTC6240 is negligible above 100 Hz. 1 1
Thus, the noise voltages are modeled here as white noise Ctot = Cpiezo + Cpar + Cf + Cgnd . (15)
2 2
sources. Recognizing that the overall amplifier will reject q
common-mode voltage noise, define From this point forward, q 2 in is referred to as the equivalent
noise charge (ENC) density.
vdiff,v,oa1 = vin−,v,oa1 − vin+,v,oa1 , (5)
1 C. Practical component selection
Zdiff = Rpar ∥ (6)
jω(Cpiezo + Cpar + Cgnd /2) Important design guidelines can be extracted from (14)
and and (15). Parasitic leakage conductance and capacitance are
1 universally bad from a noise perspective, and should be
Zf = Rf ∥ , (7)
jωCf minimized for any given sensor design. Minimizing parasitic
capacitance is especially important, as the 2voa1 2 C2
where vin+,v,oa1 and vin−,v,oa1 are the noise voltages at vin+ tot term
and vin− , respectively. Then, vdiff,v,oa1 drives the internal in (14) is a significant part of the amplifier noise floor.
voltage Furthermore, the ratio of Cpiezo to Cf is effectively the voltage
Zdiff + 2Zf gain of the first stage; Cf should be several times smaller
vint,v,oa1 = vdiff,v,oa1 , (8)
Zdiff than Cpiezo to minimize the second-stage contribution to the
noise floor. We have built working prototypes with Cf up to
with corresponding input-referred noise current
10 pF, but those with Cf = 1 pF work quite well. Since the
jω Zdiff + 2Zf differential charge amplifier requires good matching between
iin,v,oa1 = · vdiff,v,oa1 . (9)
Gint Zdiff the two input stages to achieve an acceptable common-mode
Finally substitution of (6) and (7) into (9), and recognition rejection ratio, we use PCB capacitors to implement Cf . By
that the two op-amp noise voltages are independent, yields using a four-layer PCB and building the capacitors between
the bottom two layers, we can implement each capacitor in a
2
Zdiff + 2Zf 3 mm × 3 mm area with good matching and shielding.
i2 in,v,oa1 = ω 2 2 v 2 oa1 , (10)
Gint Zdiff The value of Rf requires more care. Ideally, Rf should be
as large as possible but increased Rf gives worse bias stability.
as the corresponding input-referred current variance density,
We observed that increasing Rf beyond 10 GΩ does not yield
where v 2 oa1 is the voltage variance density of oa1.
significant performance benefits.
Each noise current of the LTC6240 can be modeled using
The centerpiece of the amplifier is the low-noise op amp
i2 oa1 = ι2 oa1 + ω 2 q 2 oa1 (11) used for the first stage, as this sets the absolute lower bound on
the noise floor. Choosing this op amp based on (14) requires
where ιoa1 and qoa1 are both constants [15]. It is further balancing vin2 and i2 over the desired frequency range and
in
assumed that voa1 , ιoa1 and qoa1 are all independent; the sensor capacitance. This requirement rules out op amps with
correlation between op amp current noise and voltage noise is bipolar or JFET input stages because these op amps typically
unspecified in [15]. When two op amps are used to construct a have unacceptably high current noise. Op amps with CMOS
differential amplifier that rejects common-mode current noise, input stages have voltage noise several times higher than
the resulting input-referred current variance density becomes top-of-the-line JFET or bipolar op amps, but with far lower
ι2 oa1 + ω 2 q 2 oa1 current noise. Of these, the LTC6240 appears to offer the
i2 in,i,oa1 = . (12) best combination of voltage noise and current noise, with
2
 6

Property LT1792 LTC6240 LTC6081 LTC6078

√ Measured Amplifier Noise Floor
v̄in (nV/√ Hz) 4.2 7 13 18
īin (fA/ Hz) 10 0.56 0.5 0.56 Measured (0 pF load)
Analytical (0 pF load)
Cin 27 pF 3.5 pF 3 pF 10 pF 10-2
Power 76 mW 6.7 mW 1.2 mW 200 µW LTspice (0 pF load)

Charge Noise [fC Hz ]

TABLE II: For small capacitance devices, the LTC6240 offers
the best performance, at the expense of power consumption
compared to the LTC6081 and LTC6078.
10-3
Measured Amplifier Gain (10 pF Series Load)

101

10-4
101 102 103 104
Gain [V/pC]

100
Frequency [Hz]
Fig. 7: The output voltage noise of the amplifier under no load
on its high gain setting (20 V/pC). The amplifier achieves
10-1 an equivalent noise charge (ENC) of roughly 3 fC over a
Analytical Differential Mode
bandwidth of 200 Hz to 20 kHz.
Measured Differential Mode
Measured Common Mode
10-2 Common Mode Rejection Ratio
101 102 103 104 105
Frequency [Hz]
60
Fig. 6: We measure gain by connecting a 10 pF capacitor in
series with each input. We achieved a charge gain of 1.91 ×
1013 V/C, within 5 % of the design gain. 50
CMRR [dB]

40
the LTC6081 and LTC6078 providing respectable performance
with lower power consumption. Previous use of the LT1792, 30
which has significantly worse current noise than the LTC6240,
caused the current noise to dominate the sensor noise floor at 20
low frequencies. See Table II for an op amp comparison.
The second-stage difference amplifier requirements are far
10
more relaxed.
√ The AD8617 has a noise floor √ of approximately
25 nV/ Hz, and so contributes 50√ nV/ Hz to vint . Each
10 kΩ resistor contributes 13 nV/ 101 102 103 104 105
√ Hz. The total noise Frequency [Hz]
contribution is therefore 56 nV/ Hz. Using (2) gives √ an
input-referred white noise contribution of only 0.028 aC/ Hz, Fig. 8: A common-mode rejection ratio of 30 to 40 dB is a
which is insignificant compared to the noise floor of the reasonable expectation for this amplifier. The spikes at 60,
complete amplifier. 180, and 300 Hz appear to be due to poor shielding and long
cable runs to test equipment.

D. Specifications
Our amplifier has a measured gain of 19.1 V/pC over a The principal reason for building a custom charge amplifier
−3 dB bandwidth of 160 Hz to 50 kHz. This comes to within is the lack of commercial low-noise amplifiers available
5 % of our 20 V/pC target gain and exceeds our minimum for low-capacitance sensors. Table III illustrates this by
target bandwidth of 200 Hz to 20 kHz. We measured an comparison. The CEC 1-328 is the highest-performing
equivalent noise charge over this target bandwidth of 30 aC commercially available differential charge amp we could find,
(185 e− ) with no sensor attached. With one of our sensors while the Femto HQA-15M-10T is the best available single-
attached, we measured the noise floor to be 62 aC (385 e− ). ended charge amp. Our amplifier outperformed both, although
Figure 6 shows the transfer function of our amplifier; Figure their datasheets did not clearly specify the test load capacitance
7 shows its noise floor while unloaded and loaded with our or spectral noise density. We also found references to charge
sensor. Note that the analytically derived noise floor closely amplifiers in the literature. The singled-ended charge amp
matches the noise floor simulated using LTspice. inside the ELectrostatic Dust Analyzer (ELDA) [14], [16] used
 7

Amplifier qnoise Cload Bandwidth

Our amplifier 30 aC 0 pF 160 Hz–20 kHz Frequency Response

Sensor Charge re Ear Canal Pressure [fC/Pa]

Our amplifier 62 aC 25 pF 160 Hz–20 kHz
CEC 1-328 [18] 500 aC N/A 5 Hz–10 kHz
Femto HQA-15M-10T [19] 350 aC N/A 250 Hz–15 MHz 102
ELDA (LTC6240) [14], [16] 57 aC 5 pF 7 Hz–10 kHz
Kelz et. al. [17] 18 aC 5.4 pF 7 Hz–300 kHz

TABLE III: A comparison of this paper’s preamp to charge Ear Canal

amplifiers available commercially and in literature. Some 100 Pinna
Noise Floor [fC]
datasheets did not specify load capacitance. Our amplifier
demonstrates good input–referred noise charge at audio
frequencies.
10-2

Reference Mic 102 103 104

LDV Window Frequency [Hz]
0

Phase [degrees]
r
eake -100
Sp ube
T -200

o pe
n d osc 102 103 104
E Frequency [Hz]

Fig. 10: The frequency response of the UmboMic appartus

relative to ear canal pressure. Extrapolated response including
Sensor Holder gain from the pinna (45◦ azimuth) is also shown. Note that
the noise floor is in units of fC and not normalized by ear
canal pressure.

Fig. 9: A 3D-printed clamp holds the sensor under the umbo.

A speaker introduces sound pressure to the sealed ear canal. Then, the Fourier transform is smoothed and normalized to a
1/3-octave bandwidth to permit direct comparison of the noise
floor and sensitivity in the same graph, as shown in Figure 10.
an LTC6240 and performed similarly to our design, while We also measure EMI sensitivity by placing the UmboMic
Kelz et. al. [17] created a fully-integrated differential charge sensor inside an aluminum foil ball without the sensor
amp with excellent noise performance. touching the foil. The foil is connected to a voltage source,
thus placing the UmboMic sensor inside a nearly uniform
IV. M EASUREMENT TECHNIQUES electric potential. Because our charge amplifier has a well-
We test our sensors in fresh-frozen cadaveric human defined charge-to-voltage gain, we can accurately compute the
temporal bones (no chemical preservatives) and conduct all “EMI capacitance” of the UmboMic sensor, namely the charge
measurements inside a soundproof and electrically isolated induced by an external potential, and hence Cgnd .
room at the Massachusetts Eye and Ear (MEE). This allows us
to take accurate measurements without background electrical, V. R ESULTS
vibrational, or acoustic noise. Fresh cadaveric human temporal A hearing device should ideally have a flat frequency
bones are procured through Massachusetts General Hospital. response from 100 Hz to 4 kHz, as this is the frequency range
Figure 9 shows our temporal bone test setup. A 3D-printed of human speech [20]. The UmboMic apparatus performs
clamp holds the UmboMic sensor under the umbo while a well between 100 Hz to 7 kHz, with the frequency response
transparent film of plastic seals the ear canal. An external determined mostly by the middle ear impedance; Figure 10
speaker introduces a sound pressure stimulus to the ear canal shows the frequency response of the UmboMic apparatus
– typically a sinusoidal sweep from 100 Hz to 20 kHz. A normalized to ear canal pressure (where the responses were
calibrated Knowles EK3103 probe-tube reference microphone confirmed to be in the linear region). Below about 1 kHz, the
measures this sound pressure stimulus, with the probe tube middle ear is spring-like and the frequency response of the
opening directly above the eardrum. We measure over a range UmboMic apparatus is flat. Above 5 to 6 kHz the mass of the
of ear canal pressure, from approximately 60 dB to 100 dB eardrum and ossicles dominates, causing umbo motion and
SPL (in the linear range). Umbo velocity at the tympanic thus sensor output to start decreasing.
membrane is measured with a laser Doppler vibrometry (LDV) The pinna and ear canal act like a horn and provide up to
beam through a clear window covering the ear canal. 20 dB of pressure gain between 2 kHz and 6 kHz [21]. The
The noise floor is measured by taking a Fourier transform cadaveric specimens that we work with no longer have the
of several seconds of amplifier noise with the sensor attached. pinna attached, but we can use the known transfer function of
 8

Equivalent Input Noise Linearity

60 103
UmboMic 0.5 kHz
UmboMic with pinna 1.0 kHz
50 Knowles mic.
102 2.7 kHz
EIN 1/3-octave BW [dB SPL]

6.4 kHz ity

40 8.1 kHz ear
in

Sensor Charge [fC]

101 fo
rL
30 ce
en
fer
0 Re
10
20

10 10-1

0 10-2
102 103 104
Frequency [Hz]

Fig. 11: Equivalent input noise (EIN) normalized to 1/3-octave 10-3

20 40 60 80 100 120
bandwidth. Ear Canal Pressure [dB SPL]
Fig. 13: No significant nonlinearity across 90 dB of dynamic
Harmonic Distortion at 1007.1 Hz @ 94.52 dB SPL range.
100
94.52 dB SPL
at 1007.1 Hz
80 reference microphone measured 33.8 dB SPL over the same
Sound Pressure [dB SPL]

frequency range.
60 Dynamic range and linearity are significant concerns for
hearing aid microphones. A frequency-domain plot of the
29.16 dB SPL UmboMic apparatus response to a 1 kHz stimulus is shown in
40 at 2014.2 Hz
Figure 12, demonstrating less than 0.1 % harmonic distortion
at 94.5 dB SPL at the eardrum. At a 114.5 dB SPL stimulus
20
level, harmonic distortion was measured to be less than 1 %.
Additionally, Figure 13 shows that the UmboMic apparatus is
0 linear across at least 80 dB of sound stimulus level.
The UmboMic apparatus also effectively rejects EMI from
-20 common sources like switched mode power supplies and
102 103 104 60 Hz mains hum. Our measured “EMI capacitance” was
Frequency [Hz] approximately 0.6 fF, which represents an improvement of
Fig. 12: Harmonic distortion of less than 0.1 % is demonstrated roughly 54 dB from our lab’s older single-ended unshielded
at 94.5 dB SPL. designs [23]. We also measured minimal interference from
60 Hz mains power and harmonics and minimal electrical
coupling between the test speaker and the sensor.
the pinna [22] to simulate the pressure gain from the outer ear
and extrapolate free field data. The dotted line in Figure 10 VI. C ONCLUSION
shows the result from including the pinna. The grey line shows A totally-implantable cochlear implant would significantly
the noise floor of the sensor in units of fC. improve the lives of users. The microphone component is one
We can compare the UmboMic apparatus to existing of the largest roadblocks to internalizing the entire system.
microphones through equivalent input noise (EIN), the level Here, we present the UmboMic, a proof-of-concept prototype
of acoustic noise due to the intrinsic electrical noise of of a PVDF-based microphone that senses the motion of the
the system. We compute EIN by dividing noise floor by umbo. We demonstrate that PVDF can work well as a sensing
sensitivity and normalizing to 1/3-octave bandwidth. The EIN material if designed as double-layered and paired with a
is a critical metric as it is related to the lowest sound that very low-noise differential amplifier. When considering the
the microphone can sense. Figure 11 shows the UmboMic effect of the pinna on performance, the UmboMic apparatus
apparatus EIN compared to that of a commercial hearing aid achieves an EIN of 32.3 dB SPL over the frequency range
microphone, the Knowles EK3103. We additionally simulate 100 Hz to 7 kHz—competitive with conventional hearing aid
the EIN when including pressure gain from the outer ear. microphones. Furthermore, the UmboMic apparatus has a flat
When accounting for this pressure gain, the UmboMic frequency response to within ∼10 dB from approximately
apparatus is competitive—we measured an A–weighted EIN 100 Hz to 6 kHz, low harmonic distortion, excellent linearity,
of 32.3 dB SPL from 100 Hz to 7 kHz. Our Knowles EK3103 and good shielding against EMI.
 9

Our prototype demonstrates the feasibility of a PVDF- [17] S. Kelz, T. Veigel, M. Grözing, and M. Berroth. A Fully Differential
based microphone. Our future goals are to re-engineer the Charge-Sensitive Amplifier for Dust-Particle Detectors. In 2018 14th
Conference on Ph.D. Research in Microelectronics and Electronics
UmboMic sensor out of biocompatible materials. We plan (PRIME), pages 13–16, July 2018.
to use conductors such as titanium or platinum for the [18] CEC Vibration Products, 1-328. Differential Charge Amplifier. Rev.
patterned electrodes, and replace the flex PCB with a version 16008.
[19] FEMTO, HQA-15M-10T. High Frequency Charge Amplifier, Feb 2012.
made in-house from biocompatible materials. Additionally, Rev. 3.
we must consider device packaging, power system, and [20] Diego Calero, Stephan Paul, Gesing André, Favio Alves, and Júlio
surgical hardware to securely hold the UmboMic apparatus Cordioli. A technical review and evaluation of implantable sensors for
hearing devices. BioMedical Engineering OnLine, 2018.
in place. While these engineering challenges are substantial, [21] Edgar A. G. Shaw. The External Ear, pages 455–490. Springer Berlin
our results demonstrate a suitable design concept for an Heidelberg, Berlin, Heidelberg, 1974.
implantable microphone which is competitive in performance [22] E. A. G. Shaw. Transformation of sound pressure level from the free field
to the eardrum in the horizontal plane. The Journal of the Acoustical
to conventional hearing-aid microphones. Society of America, 56(6):1848–1861, 08 2005.
[23] Benjamin G. Cary, John Z. Zhang, Christopher I. McHugh, Ioannis
VII. ACKNOWLEDGEMENTS Kymissis, Elizabeth S. Olson, Hideko Heidi Nakajima, and Jeffrey H.
Lang. An umbo microphone for fully-implantable assistive hearing
Kurt Broderick’s (MIT.nano) and Dave Terry’s (MIT.nano) devices. IEEE Sensors Journal, 22(22):22161–22168, 2022.
expertise were instrumental in designing the UmboMic’s
fabrication process. Many thanks to Yew Song Cheng (MEE,
UCSF) for helping carry out temporal bone experiments at
Mass. Eye and Ear.

R EFERENCES
[1] Aleksandar Vujanic, R. Pavelka, N. Adamovic, C. Kment, S. Mitic,
W. Brenner, and G. Popovic. Development of a totally implantable
hearing aid. In 2002 23rd International Conference on Microelectronics.
Proceedings (Cat. No.02TH8595), volume 1, pages 235–238 vol.1, 2002.
[2] Ping Huang, Jun Guo, Cliff A. Megerian, Darrin J. Young, and Wen H.
Ko. A laboratory study on a capacitive displacement sensor as an implant
microphone in totally implant cochlear hearing aid systems. In 2007 29th
Annual International Conference of the IEEE Engineering in Medicine
and Biology Society, pages 5691–5694, 2007.
[3] Envoy Medical. Acclaim cochlear implant.
[4] Medical Device Network. Envoy medical gets fda ide approval to begin
acclaim cochlear implant trial, 2022.
[5] NIH Clinical Trials. Feasibility of the mi2000 totally implantable
cochlear implant in severely to profoundly deaf adults. (tici), 2022.
[6] Audiology Worldnews. Med-el gets right under the skin with tici
implant, 2020.
[7] Application of polyvinylidene fluoride (pvdf) as a biomaterial in medical
textiles. In S.C. Anand, J.F. Kennedy, M. Miraftab, and S. Rajendran,
editors, Medical and Healthcare Textiles, Woodhead Publishing Series
in Textiles, pages 342–352. Woodhead Publishing, 2010.
[8] Sonion, 65GG31T. Miniature electret condenser microphone for hearing
instruments. Rev. 2, 15 September 2015.
[9] Knowles Acoustics, EK3103. EK Series Microphones.
[10] R.J. Wood, E. Steltz, and R.S. Fearing. Optimal energy density
piezoelectric bending actuators. Sensors and Actuators A: Physical,
119(2):476–488, 2005.
[11] Aaron Yeiser. A fully-implantable low-noise emi-resistant piezoelectric-
polymer microphone and amplifier for the middle ear. Master’s thesis,
Massachusetts Institute of Technology, 2022.
[12] Jayant Sirohi and Inderjit Chopra. Fundamental understanding of
piezoelectric strain sensors. Journal of intelligent material systems and
structures, 11(4):246–257, 2000.
[13] Y. Hu, J.L. Solere, D. Lachartre, and R. Tutchetta. Design and
performance of a low-noise, low-power consumption CMOS charge
amplifier for capacitive detectors. IEEE Transactions on Nuclear
Science, 45(1):119–123, February 1998. Conference Name: IEEE
Transactions on Nuclear Science.
[14] N. Duncan, Z. Sternovsky, E. Grün, S. Auer, M. Horanyi, K. Drake,
J. Xie, G. Lawrence, D. Hansen, and H. Le. The Electrostatic Lunar
Dust Analyzer (ELDA) for the detection and trajectory measurement of
slow-moving dust particles from the lunar surface. Planetary and Space
Science, 59(13):1446–1454, October 2011.
[15] Analog Devices, LTC6240/LTC6241/LTC6242. Single/Dual/Quad
18MHz, Low Noise, Rail-to-Rail Output, CMOS Op Amps. Rev.
624012fe.
[16] J. Xie, Z. Sternovsky, S Auer, K. Drake, E. Grün, M. Horanyi, H. Le,
and R. Srama. Laboratory testing and data analysis of the Electrostatic
Lunar Dust Analyzer (ELDA) instrument. Planetary and Space Science,
89:63–70, December 2013.