RECENT ADVANCES in SIGNAL PROCESSING, ROBOTICS and AUTOMATION

LQR OPTIMIAL CONTROL FOR MICROMACHINED TUNNELING

ACCELEROMETER

H.E.A. Ibrahim
Electrical and Ccomputer Control Engineering Department AAST and Maritime Transport
EGYPT

Abstract:- The micromachined tunneling accelerometer has wide applications in the

automotive industry. The most important problems that face such accelerometers are range,
bandwidth, sensitivity and accuracy. This paper presents new technique for improving
system behavior, by using optimal LQR controller. Adequate desired performance is obtained
by using the proposed technique.

Key-Words:- MEMS, LQR, Optimal Control, Accelerometer, Tunneling Mechanism.

1. Introduction sensors has emerged recently that, it

All accelerometers share a basic utilizes a constant tunneling current
structure consisting of an inertial mass between one tunneling tip (attached to a
suspended from a spring [1-8]. However, movable microstructure) and its counter
they differ in sensing the relative position electrode to sense deflection. Owing to the
of the inertial mass as it displaces under exponential relation of the tunneling
the effect of an externally applied current to the tunneling distance, these
acceleration. A common sensing method is microsensors potentially have a small
the measure of the variable capacitance device area, high sensitivity, and wide
between the two plates, where the mass bandwidth with very simple readout
forms one side of a two-plate capacitor and electronics [2,3]. A number of
the other plate is fixed. This approach micromachined tunneling sensors have
requires the use of special circuits to detect been developed for the high-resolution and
minute changes in capacitance and to wide-bandwidth applications. However,
translate them into an amplified output high-voltage operation (a few 10 V or even
voltage. Another common method uses higher) is required for these devices, thus
piezoresistors to sense the internal stress limiting their applications.
induced in the spring. The primary The operating principle of the
specifications of an accelerometer are micromachined tunneling accelerometer is
range, often given in G the earth’s simple. As the tip is brought sufficiently
gravitational acceleration (1 G = 9.81 close to its counter electrode (within a few
m/s2), sensitivity (V/G), resolution, angstroms) using electrostatic force
bandwidth (Hz), cross-axis sensitivity, and generated by the bottom deflection
immunity to shock. The range and electrode, tunneling current (Itun) is
bandwidth required vary significantly, established and remains constant if the
depending on the application. tunneling voltage (Vtun) and distance
Accelerometer for airbag crash sensing are between the tip and counter electrode are
rated for a full range of ± 50 G and a unchanged. Once the proof mass is
bandwidth of about 1 kHz. In contrast, displaced due to acceleration, the readout
devices for measuring engine knock or circuit responds to the change of the
vibration have a range of about 1 G, but tunneling current and adjusts the bottom
most resolve small accelerations (< 100 deflection voltage (Vo) to move the proof
μG) over a large bandwidth (> 10 kHz) mass back to its original position, thus
[1]. A new category of micromachined maintaining a constant tunneling current.

 
 
ISSN: 1790-5117 40 ISBN: 978-960-474-157-1
 RECENT ADVANCES in SIGNAL PROCESSING, ROBOTICS and AUTOMATION

Acceleration can be measured by reading Ab

out the bottom deflection voltage in this
S o = ρ p t p hb            (2) 
2ε o kX o
closed-loop system. The top deflection
+ 1 X
electrode (Vi applied) is used to simulate mp 2
- m p s + bs + k
acceleration electrostatically for self-test
and protect the accelerometer against Ao
acceleration over range [4, 5]. ε o AbVo
a 
hb2 s αItun φ
For best performance, tunneling- +1
ωo
based devices need to be operated in a
closed loop mode [4,5], the reasons are as Fig.1 Closed loop diagram for the overall
follows: 1) the ratio of the change in the system
tunneling current to change in the
tunneling distance is extremely large, thus Where a is the input acceleration,
limiting their measurement range if they X is the proof mass displacement, mP is
are operated in an open loop mode and 2) the proof mass, k is the spring constant, b
the tunneling barrier height for two is the damping coefficient, ωo is the cut off
tunneling electrodes may vary by one frequency for the current to voltage
order of magnitude in air [1,5], thus amplifier in rad/sec, Ao is the current to
affecting open loop device sensitivity voltage amplifier dc gain, α is a tunneling
directly, which is undesirable. In closed mechanism constant 1.025 Ǻ-1 eV-1/2 , and
loop operation, the movable structures are Φ is the tunneling barrier height for two
virtually stationary so that a large distance tunneling electrodes, hb is the distance
to current gain does not limit the between the proof mass and bottom
measurement range of tunneling based deflection electrode, Ab the electrode area,
devices. Furthermore, closed loop device εo is the permittivity of the free space,
sensitivity is determined only by the where the media between the two
electrostatic feedback force rather than the electrodes is the air and Vo is the applied
tunneling barrier height [4, 5]. voltage on the plates to generate and
control the electrostatic force to bring the
2 Accelerometer Analysis and proof mass back to its original position. ρp
Mathematical Model and tp are the material density and
A micromachined tunneling thickness of the proof mass, respectively,
accelerometer system presented in [6, 7] and Xo is the initial distance between the
consists of four subsystems, suspended two tunneling electrodes.
proof mass; tunneling based mechanism,
current to voltage amplifier, and Table 1: Design Parameters for the
electrostatic feedback, the overall Tunneling Accelerometer
mathematical model can be developed as Nominal Design Parameters Values
follows. mp 5.576 μg
The above subsystems can be k 4.42 N/m
summarized as in Fig.1 the transfer b 1.097 * 10-3
function of the closed loop sensor system N.S/m (air)
can be expressed as in equation (1). fo 5 kHz
s
+1 KT Ao K FB
Loop gain 5
X(s) ωo
G(s) = = (1) k
a(s) mp 3 b 2 k KT Ao K FB 22.1 N/m
s +(mp + )s +(b + )s + k + KT KFBAo
ωo ωo ωo In addition, it is assumed that the
ε o AbVo proof mass and bottom deflection
Where, KT = αI tun φ K FB =
hb2 electrode have the same area Ab. Table (1)
ωo = 2πf o

 
 
ISSN: 1790-5117 41 ISBN: 978-960-474-157-1
 RECENT ADVANCES in SIGNAL PROCESSING, ROBOTICS and AUTOMATION

presents the nominal design parameters for with the spring constant. When the
the accelerometer taken from [6]. spring constant of the device is
increased by a large amount,
2.1 System Results and Simulation simulation results show that, the
Intensive simulation has been done, bandwidth of the overall system goes
the results obtained from the simulation down until it becomes equal to the
are declared as follows. The system has bandwidth of the current-to-voltage
three poles at (- 31542, -35 ± 2176i), and a amplifier.
zero at –31542, which assures the system
stability. 10000
Bandwidth versus mass

Fig. 2 shows the step response for 9000

the system, it reaches 4 % from the 8000

reference value at steady state, which 7000

bandwidth rad/sec
means 96% error. The step input here 6000

represents the system acceleration input (to 5000

be sensed) and the system output is the 4000

mass displacement, which will be 3000

measured, by means of the variations in

2000

1000
the tunneling current, to give the 0
accelerometer sensor reading. Based on 0 1 2 3
mass kg
4 5
x 10
6
-4

investigating system response in time and Fig.3 Bandwidth versus accelerometer

frequency domains and the above mass
equations, we can find that. 3. An increase in the bandwidth of the
current-to-voltage amplifier does not
affect the accelerometer sensitivity,
Step response
0.08

0.07 where it increases the bandwidth of the

0.06 closed loop system significantly (it
0.05
brings the system from an overdamped
to underdamped state), one can find
Displacement

that. The closed loop frequency

0.04

0.03
response, for 50 kHz bandwidth current
0.02
to voltage amplifier, the closed loop
0.01 system bandwidth is about 28 kHz.
0
0 0.05 0.1 0.15 0.2
And when the current to voltage
Time
amplifier bandwidth has been reduced
Fig.2 System step response to 1 Hz, the closed loop system
bandwidth became 100 kHz.
1. Accelerometer has sensitivity 4. By increasing the loop gain KT Ao K FB ,
proportionally increases with the proof k
mass thickness as well as the square the system goes to unstable state,
root of the proof mass area. But from which causes large increase in the
the simulation it concludes that, the closed loop system bandwidth.
closed loop system bandwidth However, the large value in the loop
decreases with large increase in the gain will make the system unstable. So
proof mass. Fig.3 shows the bandwidth that, it needs to be controlled carefully.
versus different mass dimensions. If the closed loop gain increased 100
2. Accelerometer sensitivity is inversely times that would increase the closed
proportional to the square root of the loop system bandwidth almost 4 times.
spring constant k, while the bandwidth The loop gain is linearly proportional
of the closed loop system increases to the dc amplifier gain (Ao). One can

 
 
ISSN: 1790-5117 42 ISBN: 978-960-474-157-1
 RECENT ADVANCES in SIGNAL PROCESSING, ROBOTICS and AUTOMATION

conclude that, accelerometer 0

, and I is l*l unit matrix
bandwidth has nonlinear relationship 0
with the closed loop gain and The control objective is to have z(t)
accelerometer mass. to track r(t) and in the same time minimize
From the above analysis and a LQ performance measure. Using the
discussion we can conclude that, the performance measure J in equation (8) can
system needs controller to assure the readily accommodate the control
optimal performance and preventing the objectives, Fig.4 shows the block diagram
system to be unstable. So, the next section for the control system.
will present a new approach for using
linear quadratic regulator (LQR) optimal (8)
controller to assure tracking the input
reference and optimum performance Where J is the optimal control performance
(accurate sensing i.e. mass moves to the measure, KI is the integral gain, Q is n*n
predetermined position according to the symmetric, positive semidefinit matrix,
input, making sure the mass gets back to
weighting matrix (Q ≥ 0), R is l*l
the original position after removing the
symmetric, positive definite weighting
input signal, faster behavior, stable system,
matrix (R > 0), P is n*n symmetric positive
oscillations reduction for any mass
semidefinit constant matrix can be found
displacement, desired bandwidth and
from riccati equation, ()T means the
sensitivity).
transpose of the matrix, (ti , tf) is the
integration period and n is the number of
3. LQR Optimal Control
the state variables, l is the number of
The micromachined tunneling
inputs.
accelerometer model is a linear time
invariant system and can be put as in D
- z(t)
equation (3). + +
u(t)
, , ∫ KI(t)
G(s) C
r(t) - y(t)
(3)
Optimal
Where x(t) is the state vector, u(t) is gain K(t) x(t)
the control input, y(t) is the observed or
measured vector, z(t) is the tracking vector P(t)
 
that is required or follow a reference vector
r(t), and A,B,C,D are constant matrices of P(0) PT
compatible dimensions. Let r(t) be a t
reference input that z(t) must track or Fig.4 Optimal feed back control system
follow, such that, the tracking error e(t) is
represented as in equation (4) and the So, in this case Q and R matrices
integral error α(t) in equation (5). could be represented as in equation (9)
0 ∞ (4) 0 0
. 0 (5) 0 0 , 1 (9)
From equation (3 and 4) let, 0 0
0 To start the solution we have to
0 0 assume R = 1, and assuming any
0 magnitudes for q11 , q22 and q33 then doing
(6)
the simulation and checking the state
Or in another form as in equation (7) variables performance to get the desired
(7) state variables performance.
0 The optimal control, which
Where ,
0 minimizes J, is given by equation (10).

 
 
ISSN: 1790-5117 43 ISBN: 978-960-474-157-1
 RECENT ADVANCES in SIGNAL PROCESSING, ROBOTICS and AUTOMATION

, (10) gives zero tracking error, which means

much better sensitivity and accuracy for
Where P(t, tf) = PT is n*n matrix the micro accelerometer compared to
can be found from riccati equation (11). previous approaches [5,6,7].
The speed of reaching the final
, , , values depends on choosing the values of
matrix Q, as choosing high values of Q as
, , (11)
having faster response for any input signal
and having better stability.
from equation (10) and (4)
Finally, for all inputs, the mass
speed when using the LQR is faster and
almost has no oscillations, for example, for
(12) unit step input the average speed is
1.824*10-3 m/sec. Compared to
Where K(t) is the optimal feedback -4
1.7111*10 m/sec in [7]. That makes, the
controller gain , . mass reaches its final position faster with
no oscillations. Using LQR has increased
4. System Stability and improved the accelerometer
Using Lyapunov indirect method bandwidth, stability, accuracy and
for linear systems, the system stability can response time.
be checked, we are going to concentrate 1.2
x versus time

our attention on the eigenvalues;

1

| | 0 (13) 0.8
Acceleration
Speed
state variables

displacement
Where λ is the eigenvalues. It is 0.6

known, for stable system, the real parts of 0.4

the eigenvalues have to be in the right side
0.2
in the S-plan. After using the optimal
control to check the stability, we replace 0

by , and use -0.2

0 1 2 3 4 5 6
the same equation (13). Time (Sec)
x 10
-5

Fig.5 State variables versus time for unit

5. Results and Simulation after Using step input tracking
LQR
-3 x versus time
Simulation using the optimal LQR 6
x 10

has been implemented. Fig. (5) shows the Acceleration

5 Speed
unit step system response. As shown from Displacement
the figure the displacement reaches its 4
final value in 3*10-6 sec., gives zero
state variables

3
tracking error and the system has better
stability. Comparing that with the results 2

have been obtained previously [5,6,7]. It is 1

quit clear using LQR has much better
response. 0

Fig. (6, 7) show unit ramp and unit -1

0 1 2 3 4 5 6
acceleration system response. As shown Time (Sec)
x 10
-3

from the figure the displacement reaches Fig.6 State variables versus time for unit
its final value in few microseconds, and ramp input tracking

 
 
ISSN: 1790-5117 44 ISBN: 978-960-474-157-1
 RECENT ADVANCES in SIGNAL PROCESSING, ROBOTICS and AUTOMATION

4
x 10
-5 x versus time
[3] J. Grade et al., “Wafer-scale
3.5
Acceleration processing, assemble, and testing of
Speed
Displacement
tunneling infrared detectors”, in Proc.
3
Int. Conf. Solid-State Sensors
2.5
Actuators, Transducers 97, Chicago,
state variables

2 IL, June 1997, pp. 1241-1244.

1.5 [4] C. Yeh and K. Najafi, “A low Voltage
1 bulk-silicon tunneling-based
0.5 microaccelerometer”, in Tech. Dig.,
0
IEEE Int. Electron Devices Meet.
(IEDM), Washington, DC, Dec.
-0.5
0 1 2 3
Time (Sec)
4 5 6
-3
1995, pp. 593-596.
x 10
[5] C. Yeh and K. Najafi, “A low voltage
Fig.7 State variables versus time for unit
tunneling based silicon
acceleration input tracking
microaccelerometer”, IEEE Trans.
Electron Device, vol. 44, no. 11, pp.
6. Conclusion
1975-1882, 1997.
Micro accelerometer analysis has
[6] C. Yeh and K. Najafi,”CMOS
been presented with optimization objective
Interface Circuitry for a Low Voltage
in mind. Micro accelerometer model is
Micromachined Tunneling
presented and simulated Using MATLAB.
Accelerometer”, JMEMS IEEE, vol.
Device Sensitivity, Bandwidth and
7m No. 1, March 1998, pp. 6-15.
stability for the micro accelerometer
[7] H.E.A.Ibrahim, “Design Parameters
closed loop system are investigated. LQR
for Micromachined Tunneling
algorithm has been presented and
Accelerometer Using Genetic
designed. The system has been simulated
Optimization”, Ein-Shams
using the LQR system. Comparing results
Engineering Journal, Vol. 40, No. 4,
(system bandwidth, stability, mass speed
pp. 787-806, Dec. 2005.
and system accuracy) among systems with
[8] H.E.A.Ibrahim, M. Zohdy, R. N.
LQR, without LQR and the previous work
Loh, T. O. Ogunyemi and J. Kobus,
has been presented. The results using the
“Modeling, Simulation And Control
LQR were much better than previous
Of Nonlinear Microengine And
work. LQR guarantees wide bandwidth,
Microresonator MEMS”, Ph.D.
good accuracy, faster response and stable
Dissertation, 2002, Oakland
system.
University, Rochester, MI, USA.
[9] Frank L. Lewis and Vassilis L.
Reference:
Syrmos, "Optimal Control", 2nd
[1] Suminto, J.T., “A Wide Frequency
edition, John Wiley & Sons, Inc.,
Range, Rugged Silicon Micro
Jan. 1995.
Accelerometer with Over Range
[10] Brian D, O. Anderson and John B.
Stops”, Proc. 9th Annual Int.
Moore, "Optimal Control Linear
Workshop on Micro
Quadratic Methods", Prentice-Hall,
Electromechanical Systems, San
Inc., 1990.
Diego, CA, Feb. 11-15, 1996, pp.
180-185.
[2] T. W. Kenny et al., “Wide-bandwidth
electromechanical actuators for
tunneling displacement transducer”,
IEEE J. Microelectromechanical
systems Vol. 3, no. 3, pp. 97-104,
1994.

 
 
ISSN: 1790-5117 45 ISBN: 978-960-474-157-1