In-plane and out-of-plane MEMS gyroscopes

based on piezoresistive NEMS detection

Stefano Dellea, Federico Giacci, Antonio Francesco Longoni, and Giacomo Langfelder

 modes are kept in quadrature oscillation [16]. The QFM

Abstract— The work presents a new design and a complete gyroscopes stability is limited by temperature effects on the
characterization of AM gyroscopes based on piezoresistive nano- non-natively-matched modes. White noise is limited by supply
gauges. The working principle and optimization criteria of in- noise (which turns into frequency noise via spring softening).
plane and out-of-plane devices, relying on double frame
Lissajous FM (LFM) gyroscopes [17] bypass the offset issue
decoupling and levered sense mode, are discussed in light of
sensitivity and resolution theoretical predictions. The by continuously modulating the relative phase between the
architecture of driving and sensing electronics is also presented. modes, and in turn the sensitivity. In other words, LFM
The reduced thermo-mechanical damping with respect to gyroscopes need again a split by an intended mismatch. As a
capacitive configurations, and the inherently high output signal consequence, the obtainable signal-to-white-noise ratio has the
lead to white noise performance in the mdps/√Hz range within an same expression as in mode-split AM gyroscopes [18]. For
area smaller than 0.35 mm2, at pressures in the mbar range. Sub-
both QFM and LFM devices, the required axisymmetric
5-ppm linearity errors within 1000 dps are also demonstrated.
structure makes the design of X- and Y-axis gyroscopes quite
Index Terms— MEMS gyroscopes, NEMS gauges, inertial challenging. Only Z-axis structures were indeed shown so far.
sensors, piezoresistive sensing, motion amplitude control. An alternative approach to reduce the impact of electronic
noise is to exploit sensing principles different from capacitive
I. INTRODUCTION sensing in AM gyroscopes. In this work, gyroscopes based on
nanoelectromechanical system (NEMS) piezoresistive sensing
T HE consumer market still demands for miniaturization,
power consumption reduction, and performance
improvement of microelectromechanical system (MEMS)
elements are presented. At gauge factors in the order of 50
[19-20], NEMS gauges have an inherent advantage of giving a
gyroscopes [1-3]. Optical image stabilization capabilities drive large output signal. This consistently reduces the impact of
the requirements on white noise [4]. Navigation drives the electronics noise. Therefore, mode-split operation can again be
requirements on long-term offset drift (bias stability) [5]. chosen to preserve stability and bandwidth, without large
In this context, most of MEMS gyroscopes are based on impact of the electronic noise. Further, the absence of nested
capacitive measurement of the angular rate through the parallel plates within the sense frame is beneficial to lower the
Coriolis force occurring (along the sense direction) on a damping of the sense mode, and the associated thermo-
suspended mass kept in motion with a controlled amplitude mechanical noise, and it is also beneficial for miniaturization.
along the drive direction [6-8]. This operation is referred to as Section 2 discusses the process flow, used to combine
amplitude-modulation (AM) mode. In mode-matched AM MEMS and NEMS structural elements, which also exploits
operation (same frequencies of drive and sense modes), the eutectic bonding for reduction of dead-areas. It also introduces
gyroscope gain is the largest one, but it is difficult to typical sensing configurations based on NEMS gauges, and
simultaneously reach low-noise, large system bandwidth and presents the design of in-plane and out-of-plane gyroscopes
long-term stability at low power consumption [9, 10]. AM exploiting this elements. Guidelines and trade-offs are
gyroscopes are thus commonly operated in mode-split discussed in light of the gain-factor and of thermo-mechanical
conditions, i.e. with an intended offset between the drive and noise contributions. Preliminary electromechanical test of the
sense frequencies. Advantages of improved stability and structures in terms of mode-split and quality factor is given in
extended bandwidth come at the cost of a reduced device gain Section 3. In light of these results, Section 4 discusses the
[11], which makes the white noise density limited by the driving and sensing circuit design, analyses the electronic
electronics to sub-10 mdps/√Hz in typical products [11-15]. noise, and estimates the achievable white noise density.
Recently, to overcome this inherent drawback of mode-split Section 5 presents the experimental tests, showing high
operation, alternative working principles were proposed, based sensitivity and ultra-high-linearity (~ 5 ppm) over a 1000 dps
on frequency modulation (FM). Quadrature FM (QFM) full-scale. Within an area of less than 0.35 mm2 per axis, white
gyroscopes detect the angular rate through the frequency noise densities in the mdps/√Hz range are obtained both on Z-
variation in a mode-matched device where drive and sense and Y-axis devices, in agreement with theoretical predictions.
A long-term stability in uncontrolled laboratory environment
The authors are with the Department of Electronics, Information
of about 2 dph at observation times of 200 s is obtained.
Technology and Bioengineering, Politecnico di Milano, Milano, 20133, Italy
e-mail: [email protected].
 Fig. 1. Schematic representation of the process flow of the M&NEMS technology used for the devices in this work. Starting from an SOI wafer (a) and using
common steps like DRIE, DUV and HF attacks (b-f), the process allows designing both micrometric (15 m thick) and nanometric (250 nm thick) layers.

μm)2 at each electrode. To avoid alignment issues between the

II. PROCESS FLOW, SENSING PRINCIPLE AND DEVICE DESIGN two wafers, each electrode bonding region takes up (68 μm)2.
The devices studied in this work exploit a combination of This occupation may be reduced in future designs to minimize
micrometric and sub-micrometric elements, together with Au- dead areas. The CAP wafer features suitable cavities for getter
Si eutectic bonding between the MEMS wafer, where the deposition and minimization of the final pressure after
devices are designed, and a CAP wafer, where metal bonding. The pressured device can be at this point tilted and
interconnections are routed. wire bonded to the carrier or to a plastic package (g).

A. Process flow B. Lever system configuration for sensing

The main steps of the so-called M&NEMS process flow The described process allows to simultaneously obtain
used to fabricate the devices of this work are sketched in Fig. monocrystalline Silicon, heavily doped beams with a cross
1 [21]. Starting from a Silicon-on-Insulator wafer with a section Ag = (250 nm)2 and a length Lg of a few μm, and 15-
crystalline Silicon thickness of 250 nm (a), the first phase μm-thick inertial masses (Fig. 2a). The beams can be exploited
consists in the definition of the heavily doped NEMS patterns. as piezoresistive sensing elements [20]. When subject to an
The NEMS thin layer is removed (b) around those regions that axial stress  that causes a beam elongation or compression
define either the gauges or vertical electrodes (e.g. for self- ΔLg, the beam resistance R vary by a quantity ΔR, according to
test, or quadrature compensation). Inside these regions, the its gauge factor GF and to the Young’s modulus E.
∆𝑅 𝜎 ∆𝐿𝑔
NEMS layer is kept separated, through Silicon oxide = 𝐺𝐹 ∙ = 𝐺𝐹 ∙ (1)
𝑅 𝐸 𝐿𝑔
deposition and selective etching (c), from the epitaxial layer
The direct application of an axial force on the gauges proves
(d), which can reach typical values of surface micromachining
challenging from a design point of view, as a 5-μm-long
processes, e.g. 15 μm in this work. A two-step deep reactive NEMS gauge has an axial stiffness of thousand N/m, which
ion etching (DRIE) defines the MEMS part as well as the
lateral trenches of eutectic bonding contacts. The DRIE does
not etch the nano-gauges thanks to the oxide protection. Oxide
is then removed, through a hydrofluoric acid (HF) attack (e).
The CAP wafer, separately processed with two layers of
metal interconnections, is now bonded to the MEMS wafer (f).
Bonding occurs between Au on the CAP and Si on the MEMS
wafer. The overall force during the eutectic bonding is about
30 kN, falling mostly on rectangular seal-rings with a width of
150 μm and an overall area of 1.2 mm2, for each of the ~ 2000 Fig. 2. (a) SEM picture of a NEMS gauge coupled to a suspended part on
one side and to a fixed anchor point on the other side [19]. (b) Sensing scheme
modules on the wafer. This leads to about 20 Ω parasitic of a NEMS gauge coupled to a lever system (top view). L1 and L2 are the
resistance for an effective eutectic bonding contact area of (14 NEMS gauge distances to the lever tip and rotational hinge.
makes it difficult to cope with typical resonance frequencies in
the order of few tens kHz [22]. When using NEMS gauges as
readout elements, it is therefore necessary to exploit a lever
system to transfer the inertial force to one gauge end, while
the other end is anchored. In the following, gyroscopes
exploiting NEMS gauges for the sense mode will be
described. Gyroscopes exploiting the NEMS gauges also for
the drive-mode pick-off are currently under investigation.
Considering the situation sketched in Fig. 2b, we can now
re-write Eq. (1) by considering the lever amplification factor
1/=(L1+L2)/L2 between the lever tip, which displaces by a
quantity y, and the gauge elongation.
∆𝑅 ∆𝐿𝑔 Γ∙𝑦 𝐿2 𝑦
= 𝐺𝐹 ∙ = 𝐺𝐹 ∙ ~𝐺𝐹 ∙ ∙ (2)
𝑅 𝐿𝑔 𝐿𝑔 𝐿1 𝐿𝑔
Note that in principle this coefficient is maximized by
setting a large gauge distance L2 from the hinge. In practice,
there are other aspects to be considered for the optimization:
- target sense-mode frequency: when changing the gauge
position L2, the stiffness of the hinge-gauge-lever system in
Fig. 2b changes. The stiffness of other springs involved in the
gyroscope sense mode (e.g. drive-sense decoupling springs)
should be changed accordingly, to hold the target frequency. A
major limit (L2,max < 23 μm) arises when the required width of
the decoupling springs drops below minimum design values
(1.7 μm for the used process);
- repeatability: the larger L2, the larger the stress on the
NEMS gauge (when keeping the sense-mode frequency
constant as described above). However, too large contributions
of the NEMS gauge to the overall sense-mode stiffness imply
a different sense mode tolerance to etching nonuniformities
with respect to the MEMS drive mode [19]. Targeting for Eq.
(2) a nominal ±5% repeatability from part to part (1-sigma Fig. 3. Different architectures for gyroscopes based on NEMS gauges: (a)
value), L2 cannot be extended beyond L2,max < 18 μm; single lever along a symmetry axis; (b) single-lever external to the proof mass;
- masks misalignments: the relative effect of misalignments (c) dual-lever along a symmetry axis. Comb-finger stators are not shown for
sake of clarity
between the NEMS and MEMS masks is larger for gauges
closer to the hinge. From this point of view a value L2 > 10 μm Three device configurations including a differential lever
matches the repeatability given above for possible masks system and complying with a doubly-decoupled architecture
misalignments up to 0.5 μm. were investigated and compared. Top views of the different
Taking into account all these considerations, the value L2 = architectures are given in Fig. 3 (the drive mode occurs along
17 μm was chosen as a compromise between the points above. the horizontal direction). Option (a) includes a single lever
positioned along one symmetry axis of the device. Five types
of springs (drive, tuning-fork, drive-Coriolis decoupling,
C. Z-axis device design Coriolis-lever decoupling and rotational hinge) are highlighted
The sensing architecture with a lever system needs to cope in the figure, as well as the gauge position. Option (b) includes
with the gyroscope design. The first consideration that drives a single lever externally placed at one device side, resembling
the architecture is that one should avoid to transfer the drive- first configurations proposed for this kind of gyroscope [23,
mode large motion to the nano-gauges. Therefore a doubly 24]. Option (c) includes separate levers for the sense frames.
decoupled architecture, with an anchored drive frame, a For a given available area, no significant differences in
Coriolis (or decoupling) frame, and the sense frame formed by obtainable sensitivity are found. On the other side, option (a)
a lever system similar to the description above is chosen. promises significantly better immunity to accelerations (and
TABLE I vibrations), both along the X-axis if compared to solution (b),
FEM RESULTS FOR ACCELERATION REJECTION OF DIFFERENT STRUCTURES and along the Y-axis if compared to both solutions (b) and (c).
Type (a) - stress Type (b) - stress Type (c) - stress Table 1 reports the results of finite element simulations (FEM)
Acceleration on the gauges on the gauges on the gauges for 100 g (gravity units) of acceleration in the three directions
100 g
75 kPa DIF 27 MPa DIF 60 kPa CM for the different topologies (large and/or differential stresses,
(X direction) highlighted in red, are the most critical ones). Option (a) also
100 g
(Y direction)
3 MPa CM 20 MPa CM 150 MPa CM maximizes symmetry and compactness, and in the end it was
100 g chosen as the preferred design option.
490 kPa CM 2.8 MPa CM 450 kPa CM
(Z direction) Irrespective of the chosen architecture, note the absence of
*DIF = differential mode. **CM = common mode. parallel plates within the sense frame. This implies a large
sensing mass and the reduction of squeeze-film damping, mentioned in the Introduction, such a target bandwidth cannot
often related to parallel-plate sensing. Both facts are positive be obtained with a mode-matched device, unless force
in terms of achievable thermo-mechanical noise. A large mass feedback is used). Fig. 4a and 4b report a sketch and a
also allows, for the same resonance frequency, the use of scanning electron microscope (SEM) picture of the Z-axis
wider springs, which is positive for immunity to process over- device. The overall area is (885x394) μm2. Fig. 4c and 4d
or under-etch, and therefore in terms of device repeatability. report the drive and sense modes FEM, nominally occurring at
The drive fingers overlap is dimensioned to guarantee a 18.35 kHz and 18.95 kHz respectively, and the inset in Fig. 4e
displacement x = 4 μm. The drive fingers gap is 1.1 μm. The is a detail of the NEMS gauges positioned along the lever
frequencies are designed at about 20 kHz, with an intended system close to the rotational hinge. Assuming a controlled
mode split Δf = Δ /2 = 600 Hz (the target bandwidth for drive motion amplitude (as described in the following of this
consumer applications is in the order of 200-300 Hz: as work), the lever tip displacement y as a function of the angular
rate variation ΔΩ can be written as [2]:
𝑦 𝑥
= . (3)
∆Ω ∆ω
Combining Eq. (2) and (3), one can obtain the gain-factor,
i.e. the single-ended resistance variation per unit angular rate
change.
∆𝑅 Γ 𝑥 Γ 𝑥
= 𝐺𝐹 ∙ ∙ 𝑅 ∙ = 𝐺𝐹 ∙ ∙ 𝜌 ∙ . (4)
∆Ω 𝐿𝑔 ∆ω 𝐴𝑔 ∆ω
For a nominal resistance value of 2 kΩ (corresponding to a
gauge length of 5 μm and a resistivity  = 2.7·10-5 Ω·m), a
factor L2/(L1+L2) = (17 μm)/(228 μm + 17 μm) = 0.069, and
a gauge factor in the order of 50, the expected sensitivity turns
out to be 1.48 Ω/(rad/s) = 0.026 Ω/(dps) for each NEMS. Note
that the sensitivity goes with one over the cross section, which
was chosen as the minimum one.
The rate noise density SΩ,tm, considering only the mechanical
noise of the sense frame Sy, and assuming a constant transfer
function Qeff/ks between sense frame force and displacement
(Qeff being the gain at a distance Δf from resonance, and ks
being the sense frame stiffness), turns out to be [2]:
Q 2
√4∙kB ∙T∙bS ∙( eff )
Sy kS 1 kB ∙T∙bS
𝑆Ω,tm = y = y = √ . (5)
x∙𝜔𝑠 mS 2
ΔΩ ΔΩ
In the equation above kb and T are the Boltzmann constant and
the absolute temperature, and ms and bs are the sense mode
effective mass and damping coefficient respectively. For the
design values ms = 3.9·10-9 kg, fs = 18.95 kHz and bs = 1.5·10-7
kg/s, the predicted rate noise density caused by thermo-
mechanical effects turns out to be 750 μdps/√Hz. The value of
the damping coefficient, here given for a nominal pressure of
1 mbar, is dominated by squeeze-film effect generated by the
springs folds and by quadrature nulling electrodes designed
within the sense frame (see again Fig. 4b).

D. Y-axis device design

The considerations drawn for Z-axis gyroscopes were
extended to the design of in-plane gyroscopes. Fig. 5a and 5b
are a schematic sketch and a SEM picture of the Y-axis
device, showing the doubly decoupled architecture with the
anchored drive frame, the Coriolis (or decoupling) frame, and
the sense levered system. The overall area is (788x401) μm2.
Fig. 5c and 5d report the drive and sense modes, nominally
occurring at 18.43 kHz and 18.83 kHz respectively, the latter
showing the out-of-plane rotation of the lever. This rotation
Fig. 4. (a) conceptual sketch (drive fingers not shown) and (b) SEM top
causes differential stresses on the gauges, located at the basis
view of the Z-axis device, with clearly visible pads for eutectic bonding (note of the MEMS part, as sketched in the inset given by Fig. 5e.
the absence of interconnections on the MEMS wafer); (c) FEM of the anti- In particular, the use of a single lever system, positioned
phase tuning-fork drive mode; (d) FEM of the levered sense mode; (e) detail along one symmetry axis was obtained by placing the gauges
of the NEMS gauges positioned close to the in-plane rotational hinge.
along the torsional beams of an out-of-plane lever. Note that in ∆𝑅 h 𝑥
= 𝐺𝐹 ∙ ∙𝑅∙ . (7)
∆Ω 2∙𝐿𝐿 ∙𝐿𝑔 ∆ω
this case, for a given rotation angle  of the Coriolis frame,
the gauge displacement ΔLg is determined a priori by the fact A short lever length LL apparently improves the sensitivity.
that the gauge is attached to the base of the micrometric Yet, it also increases the sense mode stiffness. As it should
structure and by the fact that the epitaxy height h is a process cope with the target frequency, in the end it trades off with the
rather than a design parameter. Defining z as the vertical minimum allowed width of the other springs taking part to the
displacement of the lever tip, with a lever length LL, Eq. (2) sense mode (drive-Coriolis decoupling in Fig. 5c). The value
and (4) becomes in this situation: of (h / 2LL) is 2.3x lower than  calculated for Z-axis devices.
h h 𝑧
∙
A mode split value of 400 Hz and a drive motion amplitude
∆𝑅 ∆𝐿𝑔 ∙𝛼 2 𝐿𝐿
= 𝐺𝐹 ∙ = 𝐺𝐹 ∙ 2 = 𝐺𝐹 ∙ . (6) of 6 μm are thus needed, for the used lever length (245 μm), to
𝑅 𝐿𝑔 𝐿𝑔 𝐿𝑔
match the gain-factor of Z-axis devices. As an alternative, the
device can be designed with the same mode-split and drive
amplitude, and the lower lever factor can be compensated in
operation, acting on the gauge bias current.
The rate noise density is affected by the 1.5 μm vertical
distance between the proof mass and the quadrature
compensation electrodes: a larger compensation area with
respect to Z-axis devices was designed, as in-plane gyroscopes
are usually more affected by skew-angle issues and associated
quadrature [25]. As a consequence the predicted damping
coefficient [2, 26] is about 10 times larger than in Z-axis
devices. According to Eq. (5), the rate noise density caused by
thermo-mechanical noise turns out to be SΩ,tm = 1.7 mdps/√Hz.
Table II summarizes the parameters of the designed devices.

III. ELECTROMECHANICAL CHARACTERIZATION

Before designing the drive oscillator and the sense interface,
a characterization was performed to verify the effective
electromechanical parameters against expected values. In
particular, the eutectic bonding, still in a process optimization
phase, required to check the values of the obtainable pressure
and quality factors. These tests were done using a modified
MCP-G from ITmems [27], enabling both capacitive and
piezoresistive readout of the MEMS modes. In the tested
devices, quadrature electrodes were used as self-test for the
sense mode actuation. Specimens were tested through a probe
station, at wafer-level in a clean room environment.
To later identify the effective pressure in each die, reference
measurements inside a vacuum probe station were first
performed on 4 Z-axis devices of a bare wafer, giving average
drive-mode Q values of 30790 at 0.1 mbar, 4080 at 1 mbar
and 520 at 10 mbar. The mode was characterized using the
ring-down response approach, as described e.g. in [19]. The
measurement results confirm that at pressures in the order of 1
mbar, expected for the final packaging, the damping range is
linear, which allows to easily estimate the effective pressure.
Similar measurements were then repeated on capped wafers
after the eutectic bonding process. Fig. 6a summarizes the
TABLE II
GEOMETRICAL DIMENSIONS AND EXPECTED SENSITIVITY OF THE DEVICES

Fig. 5. (a) conceptual sketch (drive fingers not shown) and (b) SEM top
view of the Y-axis device, with clearly visible pads for eutectic bonding (note
the absence of interconnections on the MEMS wafer); (c) FEM of the anti-
phase tuning-fork drive mode; (d) FEM of the levered sense mode; (e) detail
of the NEMS gauges positioned close to the out-of-plane rotational hinge.
 supply. This saturation represents the nonlinearity that sets the
loop-gain to 1 after the initial start-up phase, to satisfy the
Barkhausen condition [28]. The feedback capacitances of the
TIA and of the G1 stage are used to avoid the open-loop-gain
pass unity at frequencies other than the resonance [29].
The signal at the output of G2, now in the form of a square
wave, is conditioned through a resistive voltage divider and a
variable gain amplifier (VGA). The VGA output is summed to
the DC voltage through the stage G3, whose output finally
drives the MEMS.
To compensate for the observed variations in the quality
factor and/or in the resonance frequency from part to part, a
secondary loop (control loop) integrates the TIA output in
Fig. 6. (a) measured drive and sense mode frequencies and (b) order to recover a signal proportional to the motion amplitude.
corresponding drive-mode quality factors on Z-axis devices. The latter As an alternative, a charge amplifier could be directly used
measurements were used, through a comparative analysis in a vacuum probe
station on bare wafers, to later identify the pressure of packaged samples. instead of the TIA as a front-end, but would lead to worse
noise performance of the drive loop due to the current noise of
results obtained on 8 Z-axis samples (all from the central
portion of the same wafer) in terms of drive and sense
frequencies. The corresponding average mode-split is 800 Hz,
slightly larger than predicted.
Fig. 6b reports the corresponding drive mode quality factors.
The average value is 1224, with a relatively large dispersion.
Reasons of such spread are under investigation. The estimated
average pressure of the packaged dies is therefore 3.3 mbar.
The wafer-level Q-factor yield and variability for 8 Y-axis
devices was the same as for Z-axis devices, as the two
gyroscope types are located in the same module. 15% lower
values of the average drive-mode quality factor are however
obtained for Y-axis gyroscopes. As these devices have a
similar drive topology in terms of driven mass, stiffness, and
comb fingers (see Table II), the decrease in quality factor is
ascribed to slide-film damping between the proof mass and
vertical electrodes (which are not present in Z-axis devices).
Average values of the mode split were found to be in the order
of 1 kHz. This partial difference with respect to FEM
predictions is under investigation for an improved design.

IV. ELECTRONIC ARCHITECTURE AND DESIGN

The driving and readout electronics is formed by 5 main
building blocks: (i) a primary oscillator built around the drive
resonator, (ii) a secondary loop for drive motion amplitude
control, (iii) a feed-through capacitance compensation circuit,
(iv) a Wheatstone bridge sensing interface with further
amplification, and (v) a lock-in based demodulation. The
electronics is implemented using board-level circuits and is
made compatible with mounting on the rate table used for the
measurements of Section V.
The overall schematic of the circuits is presented in Fig. 7a.

A. Drive circuit oscillator

The first three blocks implement the drive circuit. The
gyroscope proof mass is kept at the ground potential and is
driven via the comb stators with an AC square wave of a few
hundred mVrms superimposed to a DC voltage of 15 V. Drive- Fig. 7. (a) schematic view of the implemented electronics for the operation
mode pick-off is performed via a second set of comb fingers. of the gyroscopes. Note the presence of a primary loop for the drive oscillator,
The output motional current im is sensed through a low-noise a control loop to set the drive amplitude, and a feedthrough compensation
circuit. The sensing interface is based on a Wheatstone bridge. Picture (b)
transimpedance (TIA) stage, further amplified through two
reports the device assembled in a ceramic package, which is then mounted on
gain stages (G1 and G2) that make it saturate to the ± 5 V the rate table for either Z-axis (c) or Y-axis (d) rate testing [1].
the chosen operational amplifier. After integration, the signal placed inside a plastic socket soldered to the board. The board
is rectified, low-pass filtered, and compared to a reference was accurately mounted on an Acutronic rate table for Z-axis
voltage Vref. At any time, the error between the actual motion and Y-axis angular rate testing (Fig. 7b to 7d). Measurements
and the reference is processed by the proportional controller to were controlled and automated through a Labview interface
drive the VGA, and in turn to adjust the AC voltage applied to that allows to implement sensitivity, bandwidth and noise
the MEMS, in order to stabilize the motion amplitude. measurements. Quadrature can be compensated according to
The operational amplifiers FTC1 and FTC2 are used to the Tatar scheme [30], but for the first measurements shown in
implement a compensation of the feedthrough capacitance. this work devices with natively low quadrature were selected.
With the assumption that the feedthrough Cft is identical in all
the devices of the same type, the compensation is based on an
initial trimming of the voltage divider (R1, R2), so that the V. EXPERIMENTAL RESULTS
compensation capacitance Cc recalls a current equal and
opposite to ift generated by the feedthrough capacitance Cft. A. Sensitivity and linearity
The implementation of this sub-circuit turns out to be helpful As mentioned in Section III, some devices used for the tests
in case the obtained package pressure is higher than expected have a mode split which differs from the design values. The
(as in this case), implying relatively low drive mode quality tested Z-axis device has the correct f ~ 600 Hz, but the tested
factors, and correspondingly high feed-through impact. Y-axis device has af ~ 1080 Hz. The devices are tested with
the same nominal drive mode amplitude x = 4 m.
The phase reference for demodulation is set by the saturated
B. Sensing interface output of the primary loop in Fig. 7. The zero-rate offset, all
The board-level sensing interface is based on a Wheatstone appearing in the quadrature channel, is of 24 dps and 16 dps
bridge: the gauges, attached to the proof mass on one side, respectively. The absence of any noticeable offsets in the
share a common contact kept to ground. The other contact is Coriolis channel is ascribed (i) to low impact of comb-drive
connected to external resistors, trimmed to the measured defects (see e.g. [23]) and (ii) to the fact the resistive readout
gauge value (~ 2 kΩ) to complete the bridge. is quite immune from possible capacitive couplings of any
As the NEMS gauges vary in opposite direction in presence signals occurring in phase with the Coriolis signal.
of angular rates, the bridge output changes accordingly. An Using the formulas of Eq. (4) and (7), and coupling them to
instrumentation amplifier (INA) with a 50x gain reads the the electronic gains of the INA, GS and lock-in stages (GINA =
bridge output, which is then adapted to the input dynamic of a 50, GGS = 2 and GLI = 10/√2), the predicted system sensitivity,
lock-in amplifier by a non-inverting, selectable-gain stage GS. normalized to the bridge voltage Vbr,
∆𝑉𝑜𝑢𝑡 1 ∆𝑅
The demodulated output of the lock-in amplifier (SRS830 = ∙ 𝐺𝐼𝑁𝐴 𝐺𝐺𝑆 𝐺𝐿𝐼 (9)
𝑉𝑏𝑟 ∙∆Ω 2∙𝑅 ∆Ω
from Stanford Research System) is digitized through a turns out to be 4.6 mV/V/dps for the Z-axis device and 1.13
Measurement Computing MCC-1608GX acquisition board. mV/V/dps for the Y-axis.
One contribution to the electronic noise generated in the The measured sensitivities turn out to be quite in agreement
sense chain is due to the Johnson noise SR of the bridge with the predictions. For the Z-axis device, Fig. 8a reports the
resistors, whose power spectral density sums up in the output voltage as a function of the Z-axis input rate up to 1000
difference implemented by the INA. The input-referred noise, dps, at 340 mV bridge biasing voltage. The best fitting curve
reported as angular rate white noise density, turns out to be: indicates 5.47 mV/V/dps. The difference (17% larger) from
S 2∙𝑅 1 the theoretical value may be ascribed to a different value of
SΩ,Jh = ∆𝑅R = ∙ √4𝑘𝐵 𝑇𝑅 ∙ ∆𝑅 . (7)
∆Ω
𝑉𝑏𝑟
∆Ω the lever factor  and/or of the NEMS gauge cross-section.
For a 2 kΩ value of the resistors, and a bridge biasing voltage Fig. 8b reports the output voltage of the Y-axis device under
of 1.8 V (as discussed later, this voltage can be duty-cycled to
save power consumption), the input-referred noise turns out to
be 491 μdps/√Hz for the gain-factor of 0.026 Ω/(dps) derived
in Section II.C and II.D .
A second contribution is given by the INA input-referred
noise SΩ,INA, which can be calculated from its voltage noise
density SV,INA as:
2∙𝑅 √2
SΩ,INA = ∙ SV,INA ∙ ∆𝑅 . (8)
𝑉𝑏𝑟
∆Ω
With the value of the used amplifier (8 nV/√Hz, INA129 from
Texas Instruments) and the same biasing voltage and gain-
factor as above, the predicted contribution turns out to be
about 950 μdps/√Hz.
The overall electronic noise value is comparable (Z-axis) or Fig. 8. Matching measured sensitivities for the Z-axis and the Y-axis
lower (Y-axis) than the thermo-mechanical noise limits, device. Due to difference between design and measured parameters (in
confirming the high potential of NEMS gauge sensing. particular the mode-split) a different biasing voltage of the bridge needs to be
used to compensate the ~6 times lower sensitivity of the Y-axis device.
The samples to be tested were mounted on ceramic carriers,
 even gyro-compassing application requirements [32], the earth
rotation rate being 15 dph.
The Y-axis device was first tested in the conditions of Fig.
8b (Vbr = 1.8 V). Three measurements and their average are
reported in Fig. 10. The obtained ARW level is in the order of
10 mdps/√Hz, compared to a predicted value of 6.3
mdps/√Hz. The measured bias stability is about 20 dph at 20 s
observation interval.
Increasing the bridge voltage leads to a reduction in the
ARW, roughly by the same factor, down to 8.2 mdps/√Hz.
The excess electronic noise is somewhat surprising as the used
board is the same as for Z-axis devices. Vertical mounting of
the board with long cables for Y-axis test (Fig. 7b) may induce
Fig. 9. Linearity error for the Z-axis device (a) and for the Y-axis device (b) a disturbance coupling, which is still under investigation.
as a percentage of the full-scale, here assumed as 1000 dps.

Y-axis angular rates. Due to the large mode split, the bridge
biasing voltage to match the same sensitivity as the Z-axis CONCLUSIONS
device turns out to be 1.8 V, resulting in a best fitting curve of The work presented novel structures of in-plane and out-of-
1.01 mV/V/dps, to be compared to a theoretical prediction of plane gyroscopes based on piezoresistive NEMS readout.
1.13 mV/V/dps. Though some differences were encountered with respect to
Fig. 9 shows the derived linearity error (% of the full-scale) FEM modes predictions, in particular for in-plane devices, the
showing unprecedented values within 5 ppm for both the obtained results demonstrate ultra-high linearity at low white
devices. To explain this result, consider that the relative noise levels within large full-scale ranges. The area taken up
linearity error for NEMS gauge readout was characterized in a by a 3-axis gyroscope of this type is lower than (1.2 mm)2.
previous work [31] to be within few %, even in single-ended For low-power applications, the current dissipated in the
configuration, for stresses as large as 2 GPa. Considering the gauges may be reduced by adopting a modulated bridge
sensitivity in terms of stress (41.6 kPa/dps) derived from Eq. voltage. Similar approaches are already exploited in other
(4), the maximum loading value (at 1000 dps rate) turns out to miniaturized sensors based on Wheatstone bridges, like
be 41.6 MPa, which is almost two orders of magnitude lower anisotropic magneto-resistance (AMR) devices [33, 34].
than the mentioned nonlinearity limit. Further characterization activities are dedicated (i) to
Measured cross-axis rejections are larger than 55 dB. The reliability tests, and (ii) to noise and sensitivity analyses under
measured bandwidth (300 Hz) is limited by the rate table environmental changes and vibrations.
mechanical response. In particular, concerning point (i) above, the mechanical
properties of the NEMS layer were characterized thoroughly
B. Noise measurements in [33]. From there estimated nominal strengths of about 6
GPa, and according to FEM results shown in Table 1 to
To verify the predicted noise performance, Allan variance analyze effects of large accelerations, one finds a failure
measurements were taken in uncontrolled laboratory stationary acceleration value along the y-axis of 0.2 Mg
environment, in different conditions of bridge biasing. The (gravity units). This value largely exceeds typical industrial
results are summarized by Fig. 10. reliability shock tests of 104 g for consumer applications [12-
The Z-axis device was first tested with a bridge voltage of 15]. Further: such large accelerations typically occur as
340 mV (corresponding to the sensitivity shown in Fig. 8a). impulsive shocks, with sub-ms duration. So taking into
Eight Allan variance curves, lasting up to 60 minutes each, account the quasi-stationary values of Table 1 is a
were obtained in three different measurement sessions, with a
good repeatability in terms of measured angle random walk
(ARW = 4.2 mdps/√Hz, dashed curves). Effects of bias
stability begin to be visible at values around 1-4 dph, and 100-
360 s (corresponding to the full acquisition time with ten-point
average in the Allan formula). In the operating conditions, the
overall ARW predicted by the noise formulas (5), (7) and (8),
accounting for the effectively measured sensitivity, is 4.1
mdps/√Hz, quite close to the measured data.
A 3-fold increase in the bridge voltage determines a
reduction in the white noise, but not by the same factor,
indicating that the dominant noise contribution begins to be
thermo-mechanical. The measured ARW value of 2 mdps/√Hz
(see the three measurements obtained in this situation,
reported in Fig. 10, circle markers) is in very good agreement
with the predicted value of 1.8 mdps/√Hz. Some instability is Fig. 10. Measured Allan variance curves in different conditions of the
visible at rate values of 1.1 dph. Such a resolution satisfies bridge biasing for the Z-axis and Y-axis devices of this work.
conservative approach. Besides, much before the MEMS sense [11] L. Prandi, C. Caminada, L. Coronato, G. Cazzaniga, F. Biganzoli, R.
Antonello, R. Oboe, A Low-Power 3-Axis Digital-Output MEMS
mass undergoes the full displacement induced by such shocks,
Gyroscope with Single Drive and Multiplexed Angular Rate Readout,
and so much before all the stress is delivered to the NEMS IEEE Int. Solid-State Circuit Conf. (ISSCC) 2011, San Francisco, Feb
gauge, the suspended MEMS part will be blocked by suitably 2011, pp. 104-106.
designed mechanical stoppers absorbing the shock, like in [12] Invensense, MPU-9150, Nine-Axis (Gyro + Accelerometer + Compass)
MEMS Motion Tracking Device. [Online]
common capacitive devices. A complete reliability
http://invensense.com/mems/gyro/mpu9150.html, 2013.
characterization campaign under shocks and drop-tests is [13] Maxim, MAX21001 Ultra-Accurate, Automotive, 3-Axis Digital Output
however still to be done. Gyroscope, product datasheet available online at
Finally, concerning point (ii) above, since this is a resistive http://datasheets.maximintegrated.com/en/ds/MAX21001.pdf, 2014.
[14] STM L3G3200D MEMS motion sensor: 3-axis digital output gyroscope,
readout, particular care will be given to characterization of
product datasheet, [Online] http://www.st.com/st-web-ui/static/active
offset and sensitivity under temperature changes. Linear /en/resource/technical/document/datasheet/DM00043564.pdf, 2014.
compensation schemes may be considered, as this is the [15] Analog Devices, ADXRS290, Ultralow Noise, Dual-Axis MEMS
approach already followed for other sensors relying on Gyroscope datasheet, [online], http://www.analog.com/, 2014.
[16] M.H. Kline, Y. Yeh, B. Eminoglu, H. Najar, M. Daneman, D. A.
resistive bridges, e.g. AMR magnetometers. With respect to
Horsley, B. E. Boser, Quadrature FM gyroscope, Proc. of IEEE MEMS
such devices however, gyroscopes operate as modulated 2013, Taipei, Taiwan, Jan 2013, pp.604-608.
sensors at 20 kHz: therefore temperature changes affecting the [17] I. I. Izyumin, M. H. Kline, Y.-C. Yeh, B. Eminoglu, C. Hyuck Ahn, V.
bridge DC output should be bypassed by the demodulation, A. Hong, Y. Yang, E. J. Ng, Thomas W. Kenny, B. E. Boser, A 7ppm,
6◦/hr frequency-output MEMS gyroscope, Proc. IEEE MEMS 2015,
and therefore offset drifts should be inherently cancelled
Estoril (Portugal), Jan 2015, pp. 33-36.
(indeed the good results shown in Fig. 10 exploited no [18] M. H. Kline, Frequency Modulated Gyroscopes, Ph. D. final
temperature compensation scheme). To accurately test effects dissertation, University of California, Berkeley, Fall 2013.
of temperature changes on gyroscope sensitivity, a mini [19] G. Langfelder, S. Dellea, A. Berthelot, P. Rey, A. Tocchio, A. Longoni,
Analysis of Mode-Split Operation in MEMS Based on Piezoresistive
climatic chamber, compatible with the rate-table setup shown
Nanogauges, IEEE Journ. of Microelectromech. Syst., vol.24, n.1,
in Fig. 7b, is under development. pp.174,181, Feb. 2015.
[20] I. Ouerghi, J. Philippe, C. Ladner, P. Scheiblin, L. Duraffourg, S. Hentz,
ACKNOWLEDGMENT T. Ernst, Nanowire Gauge Factor Extraction Method for Material
Comparison and in-Line Monitoring, Proc. IEEE MEMS 2015, Estoril
The author acknowledge E. Brigo, G. Schierano and F. (Portugal), Jan 2015, pp. 361-364.
Maspero from Politecnico di Milano for helping with models, [21] P. Robert, V. Nguyen, S. Hentz, L. Duraffourg, G. Jourdan, and J.
Arcamone, M&NEMS: A new approach for ultra-low cost 3D inertial
and A. Berthelot and P. Rey from CEA-Leti for helping with sensor, Proc. IEEE Sens. Conf., New Zealand, Oct 2009, pp. 963–966.
fabrication. The work is supported under European Union, [22] J.-T. Liewald, B. Kuhlmann, T. Balslink, M. Trachtler, M. Dienger, Y.
FP7-ICT grant 288318 (NIRVANA project). Manoli, 100 kHz MEMS Vibratory Gyroscope, IEEE Journal of
Microelectromechanical Systems, vol.22, n.5, pp. 1115-1125, Oct 2013.
[23] A. Walther, C. Le Blanc, N. Delorme, Y. Deimerly, R. Anciant,
J.Willemin, Bias Contributions in a MEMS Tuning Fork Gyroscope,
REFERENCES IEEE Journ. of Microelectromech. Syst., vol.22, n.2, Apr 2013, pp. 303.
[24] Z. Song, X. Chen, S. Huang, Y. Wang, J. Jiao, X. Li, A high-sensitivity
[1] S. Dellea, F. Giacci, A. Longoni, P. Rey, A. Berthelot, G. Langfelder, piezoresistive gyroscope with torsional actuation and axially-stressed
Large Full Scale, Linearity and Cross-Axis Rejection in Low-Power 3- detection, proc. IEEE Sensors, 2003, pp.457-460, 22-24 Oct. 2003.
Axis Gyroscopes Based on Nanoscale Piezoresistors, Proc. IEEE MEMS [25] M. S. Weinberg, A. Kourepenis, Error Sources in In-Plane Silicon
2015, Estoril (Portugal), Jan 2015, pp. 37-40. Tuning-Fork MEMS Gyroscopes, Journal Of Microelectromechanical
[2] V. Kempe, Gyroscopes, Chapter 8 in Inertial MEMS Principles and Systems, Vol. 15, No. 3, pp. 479, June 2006.
Practice, Cambridge University press, 2011. [26] A. Frangi, G. Laghi, G. Langfelder, P. Minotti, S. Zerbini, Optimization
[3] C. Acar, A. Shkel, MEMS Vibratory Gyroscopes, MEMS Reference of Sensing Stators in Capacitive MEMS Operating at Resonance, IEEE
Shelf, Springer, 2009. Journ. of Microelectromech Syst., doi:10.1109/JMEMS.2014.2381515.
[4] R. Antonello, R. Oboe, D. Pilastro, S. Viola, K. Ito, A. Cenedese, IMU- [27] ITMEMS s.r.l, MEMS Characterization Platform, Product Data-Sheet,
based image stabilization in a HSM-driven camera positioning unit, [Online]. Available: http://www.itmems.it. Accessed Jan 2015.
IEEE Int. Conf. on Mechatronics (ICM), 2013, Feb 2013, pp. 156,161. [28] J. A. Geen, S. J. Sherman, J. F. Chang, S. R. Lewis, Single-Chip Surface
[5] D. Senkal, E.J. Ng, V. Hong, Y. Yang, C.H. Ahn, T.W. Kenny, and Micromachined Integrated Gyroscope With 50°/h Allan Deviation, IEEE
A.M. Shkel, Parametric drive of a toroidal mems rate integrating Journal of Solid-State Circuits, vol. 37, N. 12, Dec 2002, pp. 1860-1866.
Gyroscope demonstrating < 20 ppm scale factor stability, Proc. IEEE [29] G. Langfelder, A. Caspani, A. Tocchio, Design Criteria of Low-Power
MEMS 2015, Estoril (Portugal), Jan 2015, pp. 29-32. Oscillators for Consumer-Grade MEMS Resonant Sensors, IEEE Trans.
[6] A. Sharma, M. F. Zaman, F. Ayazi, A 104-dB Dynamic Range on Industrial Electronics, vol. 61, n. 1, Jan 2014, pp. 567-574.
Transimpedance-Based CMOS ASIC for Tuning Fork Microgyroscopes, [30] E. Tatar, S.E. Alper, T. Akin, Quadrature-Error Compensation and
IEEE Journ. Solid-State Circ., vol. 42, n. 8, Aug 2007, pp. 1790-1802. Corresponding Effects on the Performance of Fully Decoupled MEMS
[7] C. Acar, A. R. Schofield, A. A. Trusov, L. E. Costlow, A. M. Shkel, Gyroscopes, Journal Of Microelectromechanical Systems, Vol. 21, N. 3,
Environmentally Robust MEMS Vibratory Gyroscopes for Automotive June 2012, pp. 656-667.
Applications, IEEE Sens. Journ. vol. 9, n. 12, Dec 2009, pp. 1895–1906. [31] S. Dellea, N. Aresi, A. Longoni, Linearity of Piezoresistive Nano-
[8] J. Raman, E. Cretu, P. Rombouts, L. Weyten, A Closed-Loop Digitally Gauges for MEMS Sensors, Procedia Engineering 87 ( 2014 ) 1469 –
Controlled MEMS Gyroscope With Unconstrained Sigma-Delta Force- 1472.
Feedback, IEEE Sensors Journal, vol. 9, n. 3, Mar 2009 pp. 297-305. [32] I.P. Prikhodko, S.A.Zotov, A. A. Trusov, A. M. Shkel, What is MEMS
[9] C. D. Ezekwe, B. E. Boser, A Mode-Matching Closed-Loop Vibratory- Gyrocompassing? Comparative Analysis of Maytagging and
Gyroscope Readout Interface with a 0.004°/s/√Hz Noise Floor over a 50 Carouseling, IEEE Journal of Microelectromechanical Systems, vol.22,
Hz Band, Int. Solid-State Circuit Conference (ISSCC), IEEE, 2008. n.6, pp.1257-1266, Dec 2013.
[10] A. Sharma, M. F. Zaman, F. Ayazi, A Sub-0.2 °/hr Bias Drift [33] Honeywell, Handling Sensor Bridge Offset, AN212 Sensor Products
Micromechanical Silicon Gyroscope With Automatic CMOS Mode- Application Note, Solid State Electronics Center,
Matching, IEEE Journ. Solid-State Circ., vol. 44, n. 5, May 2009, pp. www.magneticsensors.com, accessed February 2015.
1593-1608. [34] W.-M. Lai, F.-M. Hsu, W.-L. Sung, R. Chen, W. Fang, Monolithic
Integration of Micro Magnetic Pillar Array with Anisotropic Magneto-
 Resistive (AMR) Structure for out-of-Plane Magnetic Field Detection,
Proc. IEEE MEMS 2015, Estoril (Portugal), Jan 2015, pp. 901-904.
[35] S. Dellea, G. Langfelder, A. F. Longoni, Fatigue in Nanometric Single-
Crystal Silicon Layers and Beams, Journal Of Microelectromechanical
Systems, in press, DOI: 10.1109/JMEMS.2014.2352792.

S. Dellea received the M.S. degree in Physics Engineering

from Politecnico di Milano, Italy, in 2011, with a Thesis about
fatigue effects in microscale Polysilicon. In December 2011 he
won a research grant on adhesion in micro and nano devices.
He is now attending his Ph.D., focusing on the design and test
of MEMS gyroscopes based on piezoresistive NEMS gauges.

F. Giacci received the M.S. degree in Electronics Engineering

from Politecnico di Milano, Italy, in 2012, with a Thesis about
design of gyroscope with nanostructured sensing elements.
He is now attending his Ph.D., focusing on the design of
characterization and operation electronics for capacitive and
piezoresistive gyroscopes.

A. F. Longoni received the M.S. degree in Nuclear

Engineering from Politecnico di Milano, Italy, in 1972. In this
university, he is currently a Full Professor of Detectors,
Microsensors and Microsystems.
His recent research is in the fields of MEMS, radiation
detectors, and material analysis instrumentation. He is the
author of about 180 publications and of some patents.
Prof. Longoni received the Accademia Nazionale dei Lincei
Prize for novel instrumentation for cultural heritage diagnosis.

G. Langfelder received the Ph.D. degree in Information

Technology in 2009 from Politecnico di Milano, Italy, where
he is now an Assistant Professor of Digital Imaging.
His research focuses on CMOS sensors, MEMS, and related
electronics. He is the author of about 80 refreed publications
and of 4 patents. In 2014, he was a co-founder of ITmems s.r.l.
Dr. Langfelder was the recipient of the Premio di Laurea
Accenture in 2005 and the Premio per la Promozione della
Ricerca Scientifica in 2011, granted by Rotary International.