Capacitive Accelerometer
Capacitive Accelerometer
net/publication/339166946
CITATIONS READS
2 1,175
1 author:
Giulio Puccioni
Università di Pisa
2 PUBLICATIONS 3 CITATIONS
SEE PROFILE
All content following this page was uploaded by Giulio Puccioni on 11 February 2020.
Abstract — This project presents the modelling and studying of a capacitive accelerometer that has been developed for
automotive applications, in particular for airbag systems. In the case of a car accident, where there are sudden and strong
accelerations, it is necessary to measure as fast as possible their intensity and direction with good accuracy and precision,
aiming to reduce the injuries severity to passengers. This can be well done with MEMS sensor, which has all the features
to best meet these needs. This kind of sensor can recognize when the acceleration crosses a threshold in order to activate
security systems.
This work has been developed for the university course of “design of sensors and microsystems”, at the University of
Pisa, valid for the second year of master’s degree in electronic engineering. It has been used COMSOL 5.4 for the study
and analysis of all cases in our interest from a mechanical and electrical point of view; in particular the main reference is
the work [1] published by Vijayakumar, Anju et al. in 2011 in which they analyzed a similar sensor for automotive
applications.
I. INTRODUCTION
THE reference work [1] analyzes a capacitive accelerometer capable of identifying positive and negative
acceleration levels along three perpendicular directions. In [1] it’s written about a 3D accelerometer, however this
project is referred to a sensor capable of working with accelerations on an XY plane, on which it was built starting
from a planar geometry. To realize this type of sensor the technological process with which modern semiconductor
integrated circuits are manufactured is exploited, thanks to which MEMS sensors are able to have dimensions in the
order of nanometers.
In case of a car accident, it is fundamental that the airbag system reacts as fast as possible from the moment of the
impact, in order to avoid possible injuries. For this reason, it is necessary that the system analyses and give a response
in the shortest possible time. Nowadays all the cars has a control unit that can do all the necessary evaluations in a
very short time, but to do so it must be supported by an equally fast sensor system and the best way to do it is to use a
MEMS, a micro electro mechanical system.
MEMS accelerometers can be made in two different ways, either capacitive, like the one mentioned in this work or
piezoresistive. The capacitive sensors are based on a moving solid body whit a certain mass that will move when
subjected to acceleration. The movement of the mass is the most important factor, because the small displacements of
the structure lead to a variation of a capacity, as there are parts, called "fingers" that move closer or farther away
depending on the direction of acceleration. These displacements are a direct effect of the acceleration on the system.
When the capacity in question reaches a certain threshold value, it means that we are in a limit condition in the case
under consideration, therefore the system will react appropriately. In this project, it has been analyzed the behavior of
the sensor when subjected to different acceleration values considering that the structure must also withstand high
amounts, indicatively up to 150g.
The requests of the project concerned, in addition to the design of the structure (chapter III), the determination of
deflection and stress as a function of accelerations in two different directions, both vertical and horizontal (chapter
IV.a and IV.b) and the dependence of deflection and stress as a function of the thickness of the structure (chapter IV.c).
fingers. From the deviation it can be traced back the acceleration value and the capacitive system is used to calculate
it. The system can be described first by analyzing it from a mechanical point of view and then by analyzing it from an
electrical point of view to see which are the relationships between the quantities at stake.
For the mechanical description of the system, reference should be made to a damping spring-mass-system, as seen
in figure 1.
A force F, generated by an external acceleration acting on the mass, m, causes a displacement x. The differential
equation describing the system response is given by equation:
𝜕2𝑥 𝜕𝑥
𝐹(𝑡) = 𝑚 2 + 𝑏 + 𝐾𝑥
𝜕𝑡 𝜕𝑡
where b is the damping coefficient and K is the spring coefficient, the stiffness. Converting this equation in his
Laplacian domain gives back the transfer function:
𝑠
𝑥̇ 𝑚 𝑠
= =
𝐹𝑠 + 𝐹𝑛 𝑠 2 + 𝑠 𝑏 + 𝐾 𝑚𝑠 + 𝑠𝑏 + 𝐾 2
𝑚 𝑚
Here Fs is the external force that causes the real displacement of the proof mass and which is in our interest to calculate
it, Fn is a force that come from a noise disturbance and could give back wrong results, however its analysis should be
deepened, but it is not the task of this work, so it is considered negligent. The relationship between force and deviation
can be derived, considering that 𝑥̇ = 𝑠𝑥:
𝑥 1
= 2
𝐹 𝑚𝑠 + 𝑠𝑏 + 𝐾
This is a simple function with two poles and could be characterized by the quality factor Q:
√𝐾𝑚 (1)
𝑄=
𝑏
And the resonance frequency ωres:
𝜔𝑟𝑒𝑠 = √𝐾/𝑚 (2)
𝑚𝑎 𝑎 (4)
𝑥= = 2
𝐾 𝜔𝑟𝑒𝑠
From (4), to have bigger deviations of x, so being able to measure it better, it is necessary to have smaller ωres, but
lowering it also reduces the system bandwidth, so is required to find a compromise.
As said, the accelerometer exploits the displacement induced by the acceleration on the mass to induce a variation
of capacity. The system is sensitive to acceleration in the XY plane and to do so it is used the same concept in two
different ways, i.e. the fact that the capacitance for a capacitor is directly proportional to a surface and inversely
proportional to the gap between the armatures, as said by the formula:
𝐴
𝐶 = 𝜖0 𝜖𝑟
𝑑
ε0 is the dielectric permeability of the vacuum and is worth 8,854∙10-12 F/m, εr is the dielectric permeability of the
material that separates the armatures, in our case it is taken as 1, considering that there will be only air, A is the total
overlap surface between the electrodes and d is the gap. In order to achieve higher signal, it should be increased the
total capacity and to do so it would be necessary to expand dimensions or the number of fingers. In [1] there are few
fingers if compared with other works, but not to change too much it has been used the same number of fingers.
In the case of this project the d will not be the same for both plates because if it were so we would not be able to
discriminate the displacements along the y-axis, that's why we made the distances d1 and d2 in a 1:2 ratio, as analysed
in [5]; doing so, as explained in a few lines, it is possible to make the capacitances as the sum of two capacitances with
two distances d different, it is better seen from figure 3. In this way the capacitance between the fingers has a different
form:
1 1 (5)
𝐶 = 𝜖0 𝜖𝑟 𝐻𝐿 ( + )
𝑑1 𝑑2
Where L in the overlap length and H is the thickness of the structure. It has always been considered negligible the
coupling capacity between the end of the tooth and the bottom of the electrode, because it has a much smaller surface
area than the alterative capabilities. It can already be deduced that when the parameter that makes the capacity change
is d1 or d2 there will not be a linear relationship between the capacity and displacement and consequently with the
acceleration, as opposed to when the parameters L or H vary.
Figure 3. Particular of the upper part of the structure where different distances between the electrodes can be seen.
4
As said, the system uses the variation in capacity to be able to trace the acceleration. There are two capacities, one
forming between the right electrodes, C1, and another between the left electrodes, C2; depending on how the structure
moves, these capacities vary and it modifies the output voltage. The latter is obtained by applying two opposite voltage,
V, to the fixed fingers and taking the output on the central structure:
𝐹 𝑚𝑎 (9)
𝑥= =
𝐾 𝐾
So, it is directly proportional to the acceleration and so is the Vout when there is an acceleration on the x-axis:
𝑚𝑎 (10)
𝑉𝑜𝑢𝑡 = 𝑉
𝐾𝐿
When an acceleration on the y-axis happens, the system do not respond in the same way as above, but the variation
in capacity depends on the displacement that makes the distances d from the fingers vary, always considering that
there are no other spurious movements. It can be better understood by looking at the figure 5. The change in capacity
will now depend on the displacement along the y-axis that it will be, for example for an upward displacement:
1 1 (11)
𝐶1 = 𝜖0 𝜖𝑟 𝐿𝐻 ( + )
𝑑1 − 𝑦 𝑑2 + 𝑦
1 1 (12)
𝐶2 = 𝜖0 𝜖𝑟 𝐿𝐻 ( + )
𝑑1 + 𝑦 𝑑2 − 𝑦
By making the appropriate calculations and considering that d2=2d1 it’s found that:
𝑑2 + 𝑑1
𝐶1 = 𝜖0 𝜖𝑟 𝐿𝐻 ( )=
(𝑑1 + 𝑦)(𝑑2 − 𝑦)
𝑑2 + 𝑑1
= 𝜖0 𝜖𝑟 𝐿𝐻 ( )
𝑑1 𝑑2 − 𝑑1 𝑦 + 𝑑2 𝑦 − 𝑦 2
So:
3𝑑1
𝐶1 = 𝜖0 𝜖𝑟 𝐿𝐻 ( 2 )
2𝑑1 + 𝑑1 𝑦 − 𝑦 2
3𝑑1
𝐶2 = 𝜖0 𝜖𝑟 𝐿𝐻 ( 2 )
2𝑑1 − 𝑑1 𝑦 − 𝑦 2
Hence it can be seen that there is a non-linear relationship between capacity variation and displacement. The variation
between the two capacities will have rather low values, which however should be intelligible for an electronic system.
We call, for simplicity:
𝜆 = 𝜖0 𝜖𝑟 𝐿𝐻
3𝑑1 3𝑑1
𝜆( 2 )−𝜆( 2 )
𝐶1 − 𝐶2 2𝑑1 − 𝑑1 𝑦 − 𝑦 2 2𝑑1 + 𝑑1 𝑦 − 𝑦 2
=
𝐶1 + 𝐶2 𝜆 ( 3𝑑1 3𝑑1
2 2 )+𝜆( 2 )
2𝑑1 − 𝑑1 𝑦 − 𝑦 2𝑑1 + 𝑑1 𝑦 − 𝑦 2
6
finger’s mass, that is M1F = 1,114∙10-18 Kg, which multiplied by the total number of the fingers, twelve, gives the total
mass of all the fingers together:
MFS =12∙M1F = 1,337∙10-17 Kg
C. Connecting beams
For the calculation of the total mass that is subjected to the acceleration, it is not negligible that of the beams
connecting the springs to the proof mass, even if in the work [1] this aspect is not given importance. There are two
types, two long ones that connect the mass to the lateral springs and two more stubby that connect it to the other two
springs.
For the long ones:
L = 750 nm
{ H = 40 nm
W = 100 nm
Whence VLB = 3∙10 m , that gives MLB = 6,96∙10-18 Kg and since there are two beams of this type, the total mass
-21 3
This kind of springs are characterized by a K, the stiffness, which has been analysed by Senturia [3] and is defined
by the following formula:
π4 EWH 3 (13)
K=
6 (2L1 )3 + (2L2 )3
E is Young's modulus for polysilicon and is worth 160GPa, not 131GPa as said in [1], W is the thickness of the spring,
H is its height, L1 and L2 are the two lengths indicated in the figure and the two points "b" and "a" of the figure are
respectively the hooks to the anchorage and to the beam that connects the spring to the rest of the structure. Arbitrarily
we designed the anchors, coloured in red, as cubes of side 100 nm which is the same size as the W of the connecting
beam, hence the equality between L1 and L2.
8
In document [1] the same formula is used to analyse the K of the springs and the authors committed an error in
doing this. In fact it is wrongly considered Was the thickness of the whole spring, i.e. the short side of the rectangle in
the picture, however Senturia clearly defines that W should be considered as the thickness of the polysilicon beam,
considering that the spring is folded on itself and will have a constant thickness [3]. So, in [1] it is considered a W of
100nm, but this dimension is clearly excessive if compared to what is then the real structure. Even if wrong, from the
calculation of the K, the L1 and L2 of the spring, which are equal, and H are made explicit. From this starting point it
was possible to size the springs from a basic model. So, the lateral springs has the following dimensions:
L1 = L2 = 550 nm
{ H = 40 nm
W = 30 nm
The springs at the top and bottom will have different sizes, due to the fact that the proof mass is not a square, but a
rectangle:
L1 = L2 = 400 nm
{ H = 40 nm
W = 30 nm
This will give us two different K's depending on the direction of acceleration. For an acceleration in x-direction it
is found K=1,87 N/m, which will then be doubled because there are two springs, so Kx =3,74 N/m. Instead for an
acceleration y-direction is calculated a total K, already considered the double, equal to Ky =9.74 N/m. In reality, as it
is analysed in another section, the calculated displacements are different from those found in the simulations, so there
are other factors to take into account when we go to calculate the K.
F. COMSOL Design
To make a drawing of the structure with the software COMSOL 5.4 it has been used the plane geometry, first making
a planar drawing and then an extrusion, adding a third dimension, H, equal for the whole structure. The drawings were
made by hand drawing all the different rectangles that are present in the structure and making the various unions
between the shapes. The final structure of the accelerometer is as shown in figure 8. The part coloured red is the part
consisting of the springs, the proof mass and the mobile fingers; it's the one that moves when it's subjected to an
external force due to acceleration. The ones in green are the anchors to which the springs are tied and will not be
moved. The last parts in blue are the fixed fingers that will form with the mobile ones the capacitance that will vary
when there are accelerations.
Figure 7.Plane geometry of the structure. Figure 8. 3D view of the whole accelerometer.
9
Figure 9. Modules used for simulations. Figure 10. Body Load 1, to define the acceleration.
A. Acceleration on x-axis
This first analysis is based on the application of a direct acceleration along the x-axis, i.e. horizontally, to see how
the structure moves, how it is stressed and how the capacities vary. In study 1 a parametric analysis is made by varying
𝑚
the global parameter "acc" from 0 to 1500 𝑠2 , which corresponds to a variation from 0 to 150 g, which would be the
𝑚
operating range of the accelerometer; clearly g is the acceleration of gravity and is worth about 9.81 𝑠2 , which is
approximated to 10. It has been analysed fifteen values of “acc”. The deviation of the structure has been analysed
using a 3D cut line passing from the centre of the structure in y direction to detect the movement of the mass along
this line.
10
It can be seen that the deviation is in the order of a few femto-meters, which is consistent with what is reported on the
reference text [1]. It can also be analysed the displacement in the y-direction, which is always analysed using a 3D cut
line but this time directing it along the x-axis while keeping the y constant in the centre of the structure:
Which can be easily neglected compared to the other because it is three orders of magnitude smaller.
It may be interesting to analyze the stiffness K of the springs when the system moves along the x-axis. In this case
it can be analyzed by knowing the formula of Senturia (13), and as it has already been said the K in this case is worth
1.87 N/m, which is then doubled because there are two springs working in parallel and it's worth 3.74 N/m, considering
only the two springs on the right and on the left, because the other two springs at the top and bottom in figure 7 cannot
be defined by the model of a K that is known. Considering the maximum case, where the acceleration is 150g, in this
case a force is induced on the structure and is:
𝐹 = 𝑚 ∙ 𝑎 = 𝑀𝑡𝑜𝑡 ∙ 150𝑔 = 𝐾 ∙ ∆𝑥
𝑀𝑡𝑜𝑡 ∙ 150𝑔
∆𝑥 = ≅ 51 𝑓𝑚
𝐾
From the simulation it is find out that:
∆𝑥|150𝑔 ≅ 12 𝑓𝑚
So, it is clear that the analytically calculated K is too small and does not correspond to that obtained from the
simulations, in fact the real K of the structure for accelerations along the x-axis would be:
𝑀𝑡𝑜𝑡 ∙ 150𝑔
𝐾𝑥−𝑟𝑒𝑎𝑙 = ≅ 16 𝑁/𝑚
∆𝑥|150𝑔
Then it is not negligible at all the contribution given by the springs at the top and bottom, which would give a
considerable stiffening to the structure.
As far as the stress to which the structure is subjected, it can be seen from image 15 that most of the stress will be
accumulated on the springs at the top and bottom and at the attachment between the proof mass and the connection
bars.
It can be analysed the stress again along both 3D cut lines used to analyse the displacement:
Figure16. Stress computed on the long x (1) and then long y (2) cut line.
It can be seen that stress never reaches high values in the accelerometer’s range of operation, so it will not cause
problems even if subjected to stronger accelerations; the highest peaks are always found near joints between the spring
connection beams and the proof mass.
With regard to the capacity, as said, it had to be calculated with two consecutive simulations in order to have the
respective Maxwell capacitances between the fixed and mobile electrodes. To do this, another block in the geometry
of the structure was used to wrap it all around to simulate the air in the structure, with a dielectric constant εR=1.
Analytically it is found that the value of C is given from (5) with:
1 1 𝐿 = 370 𝑛𝑚
𝐶0 = 𝜖0 𝜖𝑟 𝐻𝐿 ( + ) → { 𝐻 = 40 𝑛𝑚
𝑑1 𝑑2 𝑑2 = 2 ∙ 𝑑1 = 40 𝑛𝑚
𝐶 = 6 ∙ 𝐶0 = 5.897 ∙ 10−17 𝐹
The factor 6 is due to the fact that we have two pairs of 3 moving fingers on the proof mass for the calculation of C1
and it is exactly the same for C2. To make a numerical computation of this capacitances, a global evaluation was
carried out and the value of Maxwell capacitance was found and reported first in a table and after that in a graph.
Figure 18. The first column shows C1 in the second C2. The graph shows the two values.
These capacities values are about twice as high as those calculated analytically with zero acceleration, which can be
considered acceptable. The main problem is that these capacitances remain constant as the acceleration increases and
in the operating range of the accelerometer it would not be seen any changes in these values and the output voltage
calculated in (10) would only have a constant value that does not vary linearly with acceleration. Actually, when no
acceleration is applied, the two capacitances should be equal, but they deviate by a certain value, and this difference
remains constant even when the "acc" parameter increases. This can be explained by the fact that the structure shifts
by a very small value compared to the other quantities at stake and therefore the capacities vary very little. In the
reference text [1] it is said that the value found for capacitances is in the order of 10-28, but this result can be considered
incorrect, as it has been seen that a similar result is obtained also in this project when there is not air to wrap the
structure.
B. Acceleration on y-axis
The arguments made in the previous paragraph can be made again for the study along the y-axis, the difference is that
the acceleration must be applied in y direction by "body load". The analysis for deviation and stress have been done
again using the 3D cut lines defined above, but using them in reverse, i.e. using the one with y=constant to evaluate
stress and displacement due to acceleration and then using the one with x=constant to evaluate spurious displacement.
So, the main deviation was calculated on the cut line of figure 13 and it is:
The displacements in component x, on the other hand, are calculated along the 3D cut line in Figure 11:
Again, it can be seen that in this direction the displacement can be easily negligible, because it is 3 order smaller than
the one in y-direction.
It is valuated again the stiffness of the structure, remembering that the K analytically calculated is worth 4,87 N/m
and then doubled, so it is 9,74 N/m:
𝐹 = 𝑚 ∙ 𝑎 = 𝑀𝑡𝑜𝑡 ∙ 150𝑔 = 𝐾 ∙ ∆𝑦
𝑀𝑡𝑜𝑡 ∙ 150𝑔
∆𝑦 = ≅ 20 𝑓𝑚
𝐾
From the simulation it is find out that:
∆𝑦|150𝑔 ≅ 24 𝑓𝑚
In this case the K calculated numerically is smaller than the one that could be obtained by analysing the deviation from
the simulations, in fact:
𝑀𝑡𝑜𝑡 ∙ 150𝑔
𝐾𝑦−𝑟𝑒𝑎𝑙 = ≅ 8,4 𝑁/𝑚
∆𝑦|150𝑔
There is an unexpected behaviour this time, as it would have been legitimate to expect the system to respond with a
greater rigidity than the one calculated analytically.
As far as stress is concerned, even in this case the considerations made above are the same, only in this case it can
be seen that the springs where more stress will be accumulated will again be the two springs above and below the
structure. However, even in this case the structure has no problems related to this factor in its operating range. Since
even in the analysis of the acceleration along the x-axis the most stressed springs were the same, this is probably due
to the fact that they are smaller than the other two lateral ones and therefore have a greater accumulation of stress and
tensions inside them.
Along the 3D cut lines there will be stress values similar to those found above.
Figure 22. Stress computed on the long x (1) and then long y (2) cut line.
With regard to the analysis of the capacitances, the same considerations made for the acceleration in the x-direction
are valid and also the tables and graphs are the same to those in figure 18. Also, in this case a change in the capacity
values cannot be measured.
C. Parametric analysis of H
The last analysis that is reported in this project is the one about the overall thickness of the structure, called H, which
has been parameterized to see how the displacement and stress in the structure vary with this dimension. H is a
parameter that influences both the K (13) of the springs, as an increase in thickness leads to an increase in stiffness,
and in the value of the total mass, as it increases the overall volume of the structure. The analysis was made only for
an acceleration along the y-axis, considering that the conclusions would be similar to those for an acceleration along
the x-axis; the acceleration applied is worth 100 g, which should be the threshold for the trigger of the airbag system.
To assess the deviation and stress, reference was made again to the 3D cut lines used previously, in figure 13 and 11.
In study 2, a parametric sweep of H from 40nm to 1040 nm is made.
From this image it can be seen that as H increases, the deviation of the system also increases. This can be explained
by saying that although H increases the stiffness of the springs with a cube term, it will also increase the total mass of
the system that is subjected to the acceleration and the increase of K is not sufficient to compensate the effect of the
force that increases with the mass. Even with a significant increase in H there are no large displacements that can give
significant changes in capacity for position detection.
16
Figure 24 shows the displacement on the x-axis, which would be the spurious displacement with respect to the main
displacement along the y-axis. Again, it can be seen that for any H value this deviation is much smaller than that along
the y-axis and can be neglected.
In figure 25 it can be seen the stress trend in the structure, always evaluated along the 3D cut lines, there are not
great differences as H varies. Also, for this simulation the stress peaks will always be located in the same points of the
previous simulations, but their value will not increase with the size. This is because to the increase of the dimensions
is true that increases the mass and consequently the force that is applied on the system, but also increases the surface
on which the internal tensions and stresses are applied, which then will have values similar to those found previously.
From the 3D stress figure, it can be analysed the stress in the springs with the side legend, which can be useful because
it is not needed to do in-depth analysis to see how much the maximum stress is worth. In this case it can be seen that
when the dimensions increase, even with H equal to 1040 nm there will never be too high a stress, but the maximum
value will be lower than the maximum stress that can be borne by silicon.
VI. REFERENCES
[1] Vijayakumar S., Vijila G., Alagappan M., Anju Gupta, "Design and Analysis of 3D Capacitive Accelerometer for
Automotive Applications", COMSOL Conference in Bangalore, 2011.
[2] Rajib Ul Alam Uzzal, Ion Stiharu, Waiz Ahmed, "Design and Analysis of MEMS based Accelerometer for
Automatic Detection of Railway Wheel Flat", World Academy of Science, Engineering and Technology 53, 2009.
[3] Stephen D. Senturia, "Microsystem Design", Kluwer Academic Publishers, 2001.
[4] Andrea Nannini, "Microsistemi", Pisa University Press, 2017.
[5] Cheng-Kuan Lu, "Wireless MEMS Accelerometer for Real-time Small Laboratory animal activity monitoring",
Department of Electrical Engineering and Computer Science, Case Western Reserve University, January 2008.