978-1-4244-1882-4/08/$25.

00 2008 IEEE
PROC. 26th INTERNATIONAL CONFERENCE ON MICROELECTRONICS (MIEL 2008), NI, SERBIA, 11-14 MAY, 2008
Micro- and Nanosystems Based on Vibrating Structures
Z. Djuri
Abstract - Invention of the scanning force microscope in
1986 and development of micro and nanofabrication technologies
yielded a new generation of vibration based miniature sensors of
physical, chemical and biological parameters, with sensitivity
which could not be achieved before. This promoted again the
research of micro and nanoelectromechanical systems (MEMS
and NEMS) in the fields that have been dominated by
semiconductor electronics for more than a half of century.
In this paper the principles of operation of the miniature
vibrating structures are given. A particular attention is given to
the fact that the mass of these structures is of the order of tens of
picograms, and that their dimensions can be in the sub-
micrometer range, where the effects of Brownian motion of
particles in the surroundings become significant. These effects are
expressed as thermomechanical noise which in most cases
determines the ultimate sensor performances. Also, adsorption
and desorption (AD) of particles (atoms, molecules) on the
surface of miniature vibrating structures generate the AD noise.
Vibrating micro- and nanostructures are important not only
for the new sensor components, but also for a multitude of other
applications. As an illustration, I will mention a new generation
of MEMS oscillators, which successfully replaces the traditional
quartz oscillators in some contemporary applications. Also, a new
generation of processors which utilize built-in miniature
mechanical structures is envisioned.
I. INTRODUCTION
The idea to exploit the basic operating principles of
atomic force microscope (AFM) for sensing devices
appeared immediately after invention of AFM [1-10].
During the past ten years or so, based on the platform of
micro- and nanocantilevers, a new generation of sensors
emerged for measurement of physical, chemical,
biophysical and biochemical parameters with very high
sensitivity, low energy consumption and high reliability.
Most of these sensors operate in the oscillation mode. This
mode establishes either as a result of an external excitation
or due to the self-oscillation effect. In both cases the
measured parameter affects the amplitude, phase or
frequency of oscillation of the vibrating structure.
The vibrating structures can be made either of one
material only or of several different materials. This fact
expands the range of application of vibrating structures.
For example, if a cantilever is made of two materials of
different coefficients of thermal expansion, a bimaterial
effect occurs (bending of the cantilever due to change of
the temperature), that can be utilized for temperature
measurement, for detection of infrared radiation [10] etc.
There has been an explosion in the use of micro/nano-
cantilevers for sensing of various biological species [3-5].
The presence of a surrounding fluid significantly influences
the vibration characteristics of such devices. For example,
the cantilever resonant frequency decreases by an order of
magnitude when it is moved from the air in a liquid, while
the Q-factor decreases by two orders of magnitude. The
detection of single molecules in a liquid is the grand goal
of microcantilever biosensors and could have a significant
impact in the field of genomics and proteomics [3].
Vibrating microstructures are used as frequency
determining components of MEMS oscillators [11]. In
modern wireless telecommunication equipment there is a
pressing need for substitution of the quartz oscillators with
the smaller MEMS oscillators, which are manufactured
using silicon technologies and thus can be integrated on the
same chip with other electronic components. This enables
further miniaturization of the electronic devices. The
commercial use of MEMS oscillators has recently begun.
During the last several years great research efforts
have been made to utilize the vibrating nanostructures as
basic components of a new generation of computers, the
so-called nanomechanical computers (NMC) [12, 13]. The
basic element of NMC is a nanoelectromechanical single
electron transistor, which contains a nanocantilever.
In the next chapter a theoretic analysis of an
oscillating structure will be given, which enables us to
describe the operating principle of sensors which are used
in the dynamic mode, in gaseous and liquid environments.
The same theory is applicable to the analysis of vibrating
structures which are used as MEMS oscillators, and also
those intended for the future nanomechanical computers.
The theory of the main noise mechanisms in vibrating
micro- and nanostructures will also be presented. It is used
for determination of the ultimate performances of such
devices. A short overview of the MEMS oscillators will be
given, including their advantages and the problems that had
to be solved during the years of development of these
components. Finally, the nanomechanical computers will
be presented as an example of another interesting and
promising application of vibrating nanostructures.
II. THEORETIC CONSIDERATION
A. Micro- and nanocantilever vibrations
Our introductory theoretic consideration will begin
with the basics of the oscillating mechanical structures
Z. Djuri is with the IHTM Institute of Microelectronic
Technologies and Single Crystals, University of Belgrade,
Njegoeva 12, 11000 Belgrade, Serbia,
E-mail: zdjuric@nanosys.ihtm.bg.ac.yu
theory. As an illustrative example, we will use a miniature
cantilever of the rectangular cross-section A=Wh and the
length L, shown in Fig. 1.
For small amplitudes, the mechanical properties of the
beam structures can be analytically described by the Euler-
Bernoulli theory [14, 15]
( ) ( ) ( ) ( )
( ) t x F
t
t x w
t
t x w
A
x
t x w
N
x
t x w
EI ,
, , , ,
2
2
2
2
4
4

=
c
c
+
c
c
+
c
c

c
c
n p
(1)
Here, w(x,t) is the flexural deflection, EI is the flexural
stiffness, A is the mass per unit length of the resonator
beam, N is the axial force, is the damping coefficient and
F is a driving force per unit length. In micromechanical
devices, very often the beam width is large, i.e. W5h, and
thus it is necessary to use E/(1-
2
) instead of E in Eq. (1),
where is the Poisson ratio of the beam material.
The solutions of Eq. (1) are usually obtained by using
the method of separation of variables [15]. Consequently,
the displacement w(x,t) of the beam can be split into time
and position dependent components, and the result can be
expressed as a sum of the motion in each mode n,
( ) ( ) ( ) x t q t x w
n n
m

= , (2)
Here
n
(x) is the mode shape function based on the
coordinate along the beam, and q
n
(t) is the time dependent
amplitude of motion for mode n. By substituting the Eq.
(2), in Eq. (1), we obtain the equation for the mode shape
function, and the time dependent classical oscillator
equation for the modal function q
n
(t). The beam deflection
in an arbitrary point x
0
(w
n
(t)=
n
(x
0
)q
n
(t)), corresponding to
the mode n, can be obtained from
( ) ( ) ( ) t F t w k dt t dw dt t w d m
n n n n n n
= + + ) / ( ) / ) ( (
2 2
(3)
where the modal parameters are: the effective mass m
n
, the
damping factor
n
=m
n

n
/Q
n
, the resonant frequency
n
, the
Q-factor Q
n
, and k
n
=
n
2
m
n
, the stiffness constant. These
parameters are functions of the parameters from Eq. (1) and
also of the particular mode shape functions obtained for the
mechanical resonator with the corresponding boundary
conditions. On the left side of the Fig. 1 the first three
modes are shown for the cases of a homogenous (SiO
2
) and
a bimaterial (SiO
2
/Ni) cantilever clamped at one end. On
the right side of Fig. 1 the corresponding oscillating modes
are shown for a SiO
2
cantilever clamped at both ends.
The dissipation of the mechanical vibration energy of
a microbeam, which determines the Q-factor, can occur due
to internal structural damping, the support loss and the
viscous losses in the surrounding fluid.
A change in any of the coefficients that stand by the
spatial or time derivatives in Eqs. (1) and (3) cause
changes of the oscillation amplitude, phase or resonant
frequency of the beam structure. The principle of operation
of MEMS/NEMS sensors with vibrating sensing element is
based on this influence.
Fig. 1 The first three oscillating modes for a cantilever clamped at
one end (left) and at both ends (right).
As a first example, we will consider the influence of
the axial force, N, applied along a beam clamped at both
ends. This is the principle of operation of vibrating pressure
sensors. This example is also important because the
following theory can be utilized for determination of both
the influence of thermomechanical stress on the resonant
frequency and the MEMS oscillator resonant frequency
fluctuations due to temperature fluctuations.
In the case of a double-clamped beam, the exact
expression for the n
th
mode resonant frequency, ensuring
that the solution for the mode shape functions satisfies the
boundary conditions, is
2 2 2
/ 1 / ) / )( 3 / 1 (
n n n
z z E L h f c p t + = , (4)
where =3NL/(Eh
2
W), and z
n
are solutions of the equation
( ) |
.
|

\
|
+ + =
2 2
/ 1 / 1 z z tgh z z tg c c , (5)
while an approximate expression for
n
=2f
n
is [24]
5 . 0 3 2 5 . 0 2 2
)) /( 1 ( ) / ))( 12 /( ( EWh NL E L h
n n n
+ = p o e .(6)
f
0
=38.4 kHz
f
1
=240.2 kHz
f
2
=673.8 kHz
f
0
=245.3 kHz
f
1
=676.1 kHz
f
2
=1327 kHz
L
W h
1MHz
SiO
2
bridge
A
m
p
l
i
t
u
d
e

[
a
.
u
]

SiO
2
/Ni
1MHz 100k 10k
SiO
2
Numeric calculations show that for the first and the
second mode
1
=4.730,
1
=0.295,
2
=7.853,
2
=0.145.
The axial force along the beam of homogenous cross-
section can be represented by the mechanical stress, , as
N=A. For low stress , the frequency change is
proportional to the stress and the squared ratio of the length
and thickness of the beam.
It is already said in the introduction that at this time a
great interest exists for development of highly sensitive
biological sensors (for detection of certain kinds of cells in
human physiological liquids and early diagnostics of
diseases) [3]. Therefore a short introduction will be given
to the theory of oscillations of microcantilevers in liquids.
At the microscale, viscous losses in fluids are
typically two to three orders of magnitude greater than
other losses. The surrounding fluid also affects the natural
frequencies of the microbeam due to so-called added mass
of the surrounding fluid to the microbeam.
In interaction with an oscillating structure, the
surrounding fluid causes two effects: inertial and
dissipative. Hence, fluidic drag force can be expressed as
t v m v F
f f f
c c + = / (7)
where v=dw/dt is the velocity, and
f
and m
f
are the
parameters modelling the dissipative and inertial effects of
the fluid, respectively. The total external force F on the
structure in the fluid can be defined as a contribution of
both the drive force F
0
and fluidic drag, so that F=F
0
-F
f
.
In vacuum F=F
0
. Hence, based on eq. (3), the mode
equation of motion in the complex domain is
| | ( ) ( ) e e e e e
0
2 2
0
/ F w m i m = + , (8)
where m is the oscillator effective mass,
0
is the resonant
frequency, and =
0
m/Q is the damping factor of the
corresponding mode. Using (7), the equation of motion can
be reorganized for fluid as [9]
| | ( ) ( ) e e e e e
0
2 2
/ F w m i m
F F F F
= + (9)
where the equivalent parameters are m
F
=m+m
f
=m(1+
f
),

F
=
0
(1+
f
)
-1/2
and Q
F
=
F
m(1+
f
)/(+
f
).
Determination of the coefficients m
f
and
f
requires
solving of the complex Navier-Stokes equations for a fluid
that surrounds the oscillating cantilevers [6, 16]. The issues
pertinent to a vibrating microbeam in a fluid can be broadly
divided into three parts: (i) those that deal with a single
microbeam vibrating in fluid, (ii) those that deal with single
microbeam vibrating close to a surface, and (iii) those that
address the hydrodynamic coupling between microbeams
in fluids or microbeams bounded by a confined liquid [9].
As an illustration, an approximate solution will be given for
incompressible fluids, obtained in [16] by curve fitting of
the numerical results for the case (i)
) 065 . 0 Re 4 . 4 )( 2 / (
3 / 2
+ =

W m
f f
e tp (10)
) 1 Re 8 . 2 )( 2 / (
2 / 1 2
+ =

W
f f
e tp (11)
Here the Reynolds number Re=
f
W
2
/(4) describes the
ratio of inertia to friction terms, and
f
and are the fluid
density and viscosity, respectively.
B. Fluctuation phenomena
Fluctuations of parameters of miniature MEMS and
NEMS vibration structures increase as their physical
dimensions decrease [15]. These fluctuations determine
both the ultimate performances of the sensors and the
minimal power required for transition between the states 0
and 1 in digital systems. In oscillating systems, the
amplitude of these fluctuations is mostly determined by the
intensity of the dissipation processes. So, for example, the
thermomechanical noise (fluctuation of the deflection)
increases for several orders of magnitude when the
cantilever is brought from vacuum into a liquid. Thus the
sensitivity of a BioNEMS sensor is significantly reduced.
In the following text the basic mechanisms that generate
the main noises in MEMS and NEMS vibrating structures
are analysed, with the objective to achieve the optimal
performance of the devices based on such structures.
1. Thermomechanical noise. If the temperature
around our mechanical oscillator is finite and if the system
is in thermodynamic equilibrium, then the mechanical
oscillator must encompass some level of random
movement. These stochastic vibrations basically form
thermomechanical (TM) noise. The magnitude of these
random vibrations depends on the damping level in the
oscillator. Namely, to avoid breaking the second law of
thermodynamics, the model of a damped harmonic
oscillator must be supplemented by a generator of noise
force with a sufficient amplitude to keep the level of
stochastic vibrations dictated by the system temperature.
Without this generator of noise force, the system damping
will stop any oscillations, which would mean that the
system is at zero temperature. This would contradict the
assumed condition of thermal equilibrium at a temperature
different from zero.
For a linear dissipative system the generalized form
of the Nyquist theorem was given by Callen and Welton in
1951 |17|. This general theorem determines the relation
between the impedance and the fluctuations of the
generalized force. The mathematical formulation of this
so-called fluctuation-dissipation theorem has the form
( ) ( )

= e e e t d T E R F , ) / 2 (
2
(12)
where R(e) is the real part of the impedance Z(e)=F/v, and
E(e,T) is the oscillator's mean energy at a temperature T
with a frequency e
( ) ( )
1
1 ) / exp( 2 ,

+ = T k T E
B
e e e e = = = (13)
At higher temperatures (k
B
T>>e, E(e,T)~k
B
T, where
k
B
is the Boltzmann constant) spectral distribution of
thermal noise force is
( ) ( ) ( ) e e e TR k Z T k df F d F
B B TM N
4 } Re{ 4 /
2 2
,
= = = .(14)
he spectral density of displacement is
( ) ( ) { } e e e Y T k S
B w
Re ) / 4 (
2
= , (15)
where Re{Y(e)} is the real part of the mechanical
admittance Y(e)=Z
-1
(e). From (3) we obtain Y(e) and
using (15) the power spectral density of displacement is
( ) | | ( )
1 2
0
2
2 2
0 0
) ))( /( 4 (

+ = Q mQ T k S
B w
ee e e e e . (16)
2. Frequency noise. The performance of the micro-
cantilever dynamic sensors depends on various noise
generation mechanisms. These mechanisms include noise
in readout circuits (where the most important noise is the
noise arising from a cantilever deflection sensor) and
intrinsic frequency cantilever noise. The recent studies [18]
have shown that the spectral density of the deflection
sensor noise can be one order of magnitude lower than the
cantilever intrinsic noise for the frequency offset (f-f
0
),
which is less than the practically used bandwidth B
(typically less than 1kHz). Due to this, we will analyse the
intrinsic cantilever noise dominated by two basic
independent noise generation mechanisms. The first
mechanism is related to the induced stress in the cantilever
due to spontaneous fluctuations of its temperature during
the heat exchange with the ambient. These stress
fluctuations generate the resonant frequency fluctuations
( ) ( )
2 4 2 2
2
2
0
2
,
/ ) 2 /(
N T i T N
T h L f f f A = A o (17)
where
T
is the coefficient of temperature expansion, f
0
is
the stress-free cantilever resonant frequency and f
i
is the
resonant frequency of the cantilever with intrinsic stress
i
.
The frequency f
i
equals
i
/(2), where
i
is given by the
expression (6), in which N=
i
Wh.
According to [15], the temperature fluctuations are
B R T k T
th th B n
1 2 2 2 2
) 1 ( 4

+ = A t e , (18)
where R
th
is the thermal resistance, and
th
is the thermal
time constant.
The second noise mechanism is related to the case
when the proposed sensor is the self-oscillating system
with positive feedback. It is well known that self-sustained
oscillators universally exhibit linewidth broadening of
varying degree in their output power spectra. This
linewidth broadening, often referred to as phase noise, is
caused by noise inherent to the oscillator and is a measure
of spectral purity of the oscillator signal. In a short form,
the physics of the phase nose is as follows. The trajectory
of the steady-state oscillation in the deflection-velocity
state space is a closed curve due to periodicity and is called
the limit cycle. In the presence of noise, the fluctuations
would remain small in the radial (amplitude) direction due
to the tendency of the state to return to the limit cycle.
Fluctuation in the direction along the limit cycle does not
experience restoring force to return the phase to its original
value. That means that, in the presence of noise, the state
point experienced Brownian motion, or the phase
undergoes diffusion process, with the diffusion constant
D

. To determine the spectral density of the phase noise, it

is necessary to solve the Langevin or the corresponding
Focker-Planck equation.
Utilizing the definition of phase noise for given
frequency offset =2(f-f
0
) and the mentioned solution of
the stochastic equation, we obtain for the phase noise [19]
1 2 2
) ) (( 2 ) (

+ A = A +
m m
e e D D (19)
The diffusion constant is given by the expression
D

=k
B
Te
0
/(A
0
2
k
eff
Q
eff
), where k
eff
is the effective stiffness
constant, A
0
is the oscillation amplitude and Q
eff
is not a
conventional quality factor. Q
eff
, defined above, is a direct
measure of the amount of noise in the oscillator as it
includes every noise source in the detector system. If a
noise source has the thermal origin, for example a thermal
vibration of the cantilever which is the most dominant
noise source in our case, Q
eff
becomes a conventional
quality factor Q.
The relation between the spectral densities of the
frequency and phase noise is <
N
2
>=()()
2
. After
integration within the bandwidth B, the power spectral
density of the frequency noise is obtained in the form
B x arctg x C f
B N
)) ( (
2
,
= A (20)
where C=D

2
/(2t
3
B) and x=tB/D

.
It is interesting to note that for large x (x>>1), i.e. for
>>D

, the well known T.R. Albrecht et al. expression

[20] for the TM frequency noise
) 2 /(
2
0 0
2
,
kQA TB k f f
B B N
t = A (21)
can be obtained from (20). This expression is universally
used for the frequency noise analysis for the AFM.
Since the two dominant intrinsic noise generation
mechanisms are mutually independent, the power spectral
density of the total frequency noise equals the sum of the
components <Af
N,T
2
> and <Af
N,B
2
>.
3. AD processes. In the case of structures of small
dimensions and mass, and such are the micro- and
nanostructures, adsorption of particles has a significant
influence on their mechanical characteristics. Due to
adsorption on vibrating micro/nanostructures, the changes
in oscillation parameters occur as a consequence of both
the added (adsorbed) mass and the change in the stiffness
constant. The AD process can be undesirable in some
MEMS and NEMS resonant structures. Fluctuations of the
number of adsorbed particles due to the random nature of
the AD process cause the change of the resonator mass and,
consequently, the unwanted parasitic changes of its
resonant frequency (AD frequency noise). Hence the AD
process adversely affects the performances of the vibrating
MEMS and NEMS structures.
We analysed first the physical adsorption of particles
of one gas. Assuming that adsorption occurs in one layer
and that the AD process in thermodynamical equilibrium
can be described by the Langmuirs isotherm, we derived
an exact expression for the spectral density of the AD
phase noise in micromechanical resonant structures. This
noise is generated by instantaneous differences in the rates
of adsorption and desorption of molecules to and from the
resonator surface, which cause mass fluctuations and
consequentially the resonator frequency fluctuations. We
used the analogy between AD processes in resonant
structures and generation-recombination processes in
semiconductors. We presented in [21] a step-by-step exact
derivation of the AD fluctuations-induced phase noise, and
here will be given only the expression we obtained for the
spectral density of normalized frequency y (y=f/f
0
)

) 1 /( ) / ( 4 /
2 2 2 2
0
2
0 2
2
0
2
W Z W ! ' m M N C f f S
a y
, (22)
where m
0
and f
0
are the mass and resonant frequency,
respectively, of the microcantilever without the adsorbed
particles, C
2
=1/<>, <>=W
0
1
exp(E
d
/RT) is the average
time the particle spends in the adsorbed state (
0
is the
adatom period of thermal vibrations normal to the surface),
=<>/(1+bp) is the AD process time constant, which
determines the rate of reaching the equilibrium value of
adsorbed mass, E
d
is desorption energy, R is gas constant, p
is the pressure, b=
S
C
1
/(C
2
N
m
), where
S
is the adhesion
coefficient, C
1
=(2SM
a
k
B
T)
-1/2
, N
m
is the maximum number
of adatoms per unit area, N
0
denotes the number of the
adsorbed particles in the stationary state, and M
a
is the
mass of a single particle. The single sideband spectral
density of phase fluctuations is S

(f)=0.5(f
0
2
/f
2
)S
y
(f), and
finally the expression for the AD fluctuations-induced
phase noise is (f)=10logS

(f) [dBc/Hz].
Our numerical results show that the AD induced phase
noise is comparable to other sources of noise in
micromechanical resonant structures and that it prevails
when the resonator dimensions are very small. Analysis of
AD induced resonator frequency fluctuations and of the
corresponding phase noise shows that there is the strong
dependence between the resonator performance and the
environmental conditions.
Since the atmosphere around the resonator is a
mixture of gases in a majority of cases, we expanded our
analysis to address the case of the simultaneous adsorption
of particles of two or more different gases on the resonator
surface [22]. The exact expressions are derived for the
power spectral density of the fluctuations of the number of
adsorbed particles for each gas from the mixture, as well as
for the total adsorbed mass fluctuation, using the analytical
Langevin approach. As an illustration of the presented
theory, we determined the adsorbed mass fluctuations on
the surface of a sensor with a silicon micro- or
nanocantilever in the atmosphere of three gases (Fig. 2).
Fig. 2 Adsorbed mass fluctuations in the case of a mixture of
three gases, whose pressures are p
1
= p
2
=10
-3
Pa, p
3
=100 Pa. The
cantilever dimensions are 200 Pm 50 Pm 2 Pm.
The theoretic models of an AD process are useful for
both qualitative and quantitative analysis of the AD noise
and also for estimation of its contribution to the total noise
in MEMS/NEMS sensors and oscillators. Based on these
models, it is possible to determine both the minimal
detectable signal and sensitivity of the micro/nanosensors,
and to perform the analysis of the influence of the adsorbed
particles number and mass fluctuations on performance of
various MEMS and NEMS devices. Thus the theory of AD
processes is useful for optimization of their parameters and
working conditions. It is concluded that the influence of
AD induced fluctuations on both oscillator and sensor
performance becomes particularly significant as the
dimensions of the structures scale down to the order of
hundreds of nanometers and lower.
The possibility of developing the method for
identification of the gases in mixtures based on the AD
noise spectral density can also be considered [22]. While
gas identification is possible in the case of a single gas
atmosphere, the number and the position of knees in the
noise spectrum (Figs. 2 and 3 a) could be misleading when
the number, type and amount of gases in the mixture are to
be determined (i.e. when identification of gases in a
mixture is considered). However, the analysis of the AD
! '
2
m
] / [ Hz kg
f [Hz]
gas 2
gas 3
mixture
gas 1
10
0
10
2
10
4
10
6
10
8
10
-2
10
-26
10
-25
10
-24
10
-23
10
-22
10
-21
10
-20
10
-19
( ) ( )
1
1 ) / exp( 2 ,

+ = T k T E
B
e e e e = = = (13)
At higher temperatures (k
B
T>>e, E(e,T)~k
B
T, where
k
B
is the Boltzmann constant) spectral distribution of
thermal noise force is
( ) ( ) ( ) e e e TR k Z T k df F d F
B B TM N
4 } Re{ 4 /
2 2
,
= = = .(14)
he spectral density of displacement is
( ) ( ) { } e e e Y T k S
B w
Re ) / 4 (
2
= , (15)
where Re{Y(e)} is the real part of the mechanical
admittance Y(e)=Z
-1
(e). From (3) we obtain Y(e) and
using (15) the power spectral density of displacement is
( ) | | ( )
1 2
0
2
2 2
0 0
) ))( /( 4 (

+ = Q mQ T k S
B w
ee e e e e . (16)
2. Frequency noise. The performance of the micro-
cantilever dynamic sensors depends on various noise
generation mechanisms. These mechanisms include noise
in readout circuits (where the most important noise is the
noise arising from a cantilever deflection sensor) and
intrinsic frequency cantilever noise. The recent studies [18]
have shown that the spectral density of the deflection
sensor noise can be one order of magnitude lower than the
cantilever intrinsic noise for the frequency offset (f-f
0
),
which is less than the practically used bandwidth B
(typically less than 1kHz). Due to this, we will analyse the
intrinsic cantilever noise dominated by two basic
independent noise generation mechanisms. The first
mechanism is related to the induced stress in the cantilever
due to spontaneous fluctuations of its temperature during
the heat exchange with the ambient. These stress
fluctuations generate the resonant frequency fluctuations
( ) ( )
2 4 2 2
2
2
0
2
,
/ ) 2 /(
N T i T N
T h L f f f A = A o (17)
where
T
is the coefficient of temperature expansion, f
0
is
the stress-free cantilever resonant frequency and f
i
is the
resonant frequency of the cantilever with intrinsic stress
i
.
The frequency f
i
equals
i
/(2), where
i
is given by the
expression (6), in which N=
i
Wh.
According to [15], the temperature fluctuations are
B R T k T
th th B n
1 2 2 2 2
) 1 ( 4

+ = A t e , (18)
where R
th
is the thermal resistance, and
th
is the thermal
time constant.
The second noise mechanism is related to the case
when the proposed sensor is the self-oscillating system
with positive feedback. It is well known that self-sustained
oscillators universally exhibit linewidth broadening of
varying degree in their output power spectra. This
linewidth broadening, often referred to as phase noise, is
caused by noise inherent to the oscillator and is a measure
of spectral purity of the oscillator signal. In a short form,
the physics of the phase nose is as follows. The trajectory
of the steady-state oscillation in the deflection-velocity
state space is a closed curve due to periodicity and is called
the limit cycle. In the presence of noise, the fluctuations
would remain small in the radial (amplitude) direction due
to the tendency of the state to return to the limit cycle.
Fluctuation in the direction along the limit cycle does not
experience restoring force to return the phase to its original
value. That means that, in the presence of noise, the state
point experienced Brownian motion, or the phase
undergoes diffusion process, with the diffusion constant
D

. To determine the spectral density of the phase noise, it

is necessary to solve the Langevin or the corresponding
Focker-Planck equation.
Utilizing the definition of phase noise for given
frequency offset =2(f-f
0
) and the mentioned solution of
the stochastic equation, we obtain for the phase noise [19]
1 2 2
) ) (( 2 ) (

+ A = A +
m m
e e D D (19)
The diffusion constant is given by the expression
D

=k
B
Te
0
/(A
0
2
k
eff
Q
eff
), where k
eff
is the effective stiffness
constant, A
0
is the oscillation amplitude and Q
eff
is not a
conventional quality factor. Q
eff
, defined above, is a direct
measure of the amount of noise in the oscillator as it
includes every noise source in the detector system. If a
noise source has the thermal origin, for example a thermal
vibration of the cantilever which is the most dominant
noise source in our case, Q
eff
becomes a conventional
quality factor Q.
The relation between the spectral densities of the
frequency and phase noise is <
N
2
>=()()
2
. After
integration within the bandwidth B, the power spectral
density of the frequency noise is obtained in the form
B x arctg x C f
B N
)) ( (
2
,
= A (20)
where C=D

2
/(2t
3
B) and x=tB/D

.
It is interesting to note that for large x (x>>1), i.e. for
>>D

, the well known T.R. Albrecht et al. expression

[20] for the TM frequency noise
) 2 /(
2
0 0
2
,
kQA TB k f f
B B N
t = A (21)
can be obtained from (20). This expression is universally
used for the frequency noise analysis for the AFM.
Since the two dominant intrinsic noise generation
mechanisms are mutually independent, the power spectral
density of the total frequency noise equals the sum of the
components <Af
N,T
2
> and <Af
N,B
2
>.
noise spectrum in gas sensors can be used in order to
distinguish between different gas mixtures.
The resulting adsorbed mass fluctuations in the case of
the mixture can be lower than in the case of the single gas
at the same pressure (Fig. 2).
Fig. 3b shows the range of both pressure of gas 1 and
the frequency where a reduction of fluctuations exists (the
pressure of the second gas is assumed to be constant). In
the meshed part the adsorbed mass fluctuations in the case
of a gas mixture decrease by an order of magnitude
compared to the case of single-gas adsorption, while the
dark shaded part denotes the area of p
1
and f where the
reduction is of the lower order.
The diagrams in Figs. 3 a and b enable determination
of the amount of additional gas, which should be added to
the present gas in order to achieve the reduction of the
adsorbed mass fluctuations. This is useful for optimization
of the working conditions of the MEMS/NEMS oscillators,
by choosing the mixture of the surrounding gases which
enables minimization of the AD and total phase and
frequency noise and better accuracy of oscillator frequency.
Fig. 3. (a) The power spectral density of the adsorbed mass
fluctuations for the two-gas mixture, assuming a constant pressure
of gas 2 (p
2
=10
3
Pa). (b) The range of both pressure of gas 1 and
the frequency where the adsorbed mass fluctuations in the case of
a gas mixture are lower than in the case of single-gas adsorption.
In the available literature only a little amount of data
can be found about the AD process parameters on the
surface of micro/nanostructures, that is useful for
quantitative determination of the influence of the AD
process on both the response and the ultimate performances
of MEMS/NEMS sensors and oscillators. In the methods
described in the literature, the data about the AD process
dynamics are obtained by measuring the resonant
frequency change in time. Such methods are applicable
only if the adsorption process is slow in comparison with
the response rate of the oscillator itself. In contrast to these,
the method for AD process parameter determination based
on time domain analysis of experimentally obtained
oscillator transient response in the presence of adsorption
[23] is also applicable to very fast processes. The method
originated from the analysis of the solutions of a
differential equation for the first oscillation mode Eq. (3).
The mass of the oscillator changes over time due to the AD
process (m(t)=m
0
+m
0
(1-exp(-t/W)), m
0
is the adsorbed
mass in steady state). It is shown that the solution of this
equation, obtained using an iterative method, contains
information about AD process kinetics, and therefore it can
be used for determination of the AD process parameters.
The experimentally obtained oscillator time response,
w
exp
(t), can be fitted with the function Cexp(-(t-t
s
)/
f,exp
) for
t>t
s
(t
s
is the moment when AD process starts), to obtain
the time constant
f,exp
, which determines the establishment
rate of the steady state in the presence of AD process.
If
f,exp
>>
o
(
o
=2Q/Z
0
is the oscillator time constant),
the AD process time constant determines the response
envelope, and equals the constant obtained by fitting
(
f,exp
). However, if
f,exp
>>
o
is not valid, the time
constant determining the AD process kinetics can be
obtained based on Fig. 4. We created a computer
simulation which performs both the curve fitting of the
function w(t) (obtained by solving motion equation) and
determination of
f,sim
, for various given values of , and for
a constant value of
o
. When and
o
are of the same order
of magnitude, and also when is one order of magnitude
lower, can be determined from the diagram, as the value
which corresponds to the experimentally obtained
f,exp
. The
smaller cantilevers (higher resonant frequency, lower
o
)
enable characterization of fast AD process kinetics (with
time constant as low as 10
-5
s) by the described method.
Fig. 4. Time constant
f,sim
(obtained by computer simulation) as a
function of the AD time constant , for three different values of
o
.
As a part of our analysis of noise in EMS/NEMS
devices, we also considered the thermomechanical (TM)
noise in the presence of an AD process [24]. The method
we utilized is based on the Onsager's regression hypothesis
[25], which enables determination of the autocorrelation
function of the deflection, using a deterministic response
w(t), as k
B
Tw(t)/F, where F is the applied perturbation. The
cosine Fourier transform of the autocorrelation function
yields the TM noise spectrum <w
2
()> which is the
experimentally relevant quantity. It is shown that the AD
process time constant can be determined by analysis of the
oscillator response in the time domain, and the same can
also be done based on the known power spectrum of the
TM noise (Fig. 5). This theory allows the investigation of
gas adsorption-desorption kinetics using nanoscale
oscillating structures. Determination of the AD process
time constant is useful for research of both AD and
catalytic processes in general, for estimation of the AD and
total noise in MEMS/NEMS sensors and oscillators, and
also for development of the methods for characterization
and, possibly, recognition of the adsorbate.
a)
t [ms]
0 2 4 6 8 10 12 14 16 18 20
'
x
(
t
)
[
n
m
]
0
0.5
0.4
0.3
0.2
0.1
-0.5
-0.4
-0.3
-0.2
-0.1
W = 10
-4
s
10
-3
s
10
-2
s
b)

x
2
(
Z
)
!
>
n
m
2
/
H
z
@
742 742.5 743 743.5 744 744.5 745
u10
5
Z >kHz@
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0
u10
-5
m0+'m0
m0
W = 10
-4
s
10
-3
s
10
-2
s
Fig. 5. ) The difference in response; b) The TM noise in a nano-
oscillator with AD induced variable mass and of a constant mass
oscillator, for three AD processes with different time constants, .
C. MEMS silicon oscillators
An electronic frequency reference or a clock in a
digital system is based on an oscillator, which is composed
of a resonant tank element and a sustaining circuit which
drives the resonant element. The characteristics of the
resonant element, the resonant frequency, quality factor,
temperature sensitivity etc., largely determine the
characteristics of the oscillator output. Difference between
reference or clock technologies is the resonator. The most
important requirement for the frequency reference is that
the frequency of the output signal should be constant over
time. According to [11] the 21
st
century begins the era
when quartz oscillators will be abandoned, for two reasons:
i) Miniaturisation. A modern microprocessor chip may
have 10 million transistors. To operate, it requires a single
quartz frequency reference that is half of size of the entire
chip! ii) Silicon compatibility. Devices that can be
manufactured with silicon technology are promoted
because they can be manufactured cheaply using the
existing silicon batch fabrication capacity, and digital
circuitry can be integrated into them directly. MEMS
resonators have been a topic of research for almost 40
years. Significant advances have occurred in the past 20
years, and commercialization efforts begun in earnest in the
last 5 years. The reason for this was a stability problem.
The stability of the signal is the primary performance
characteristic of a frequency reference. Stability can be
classified into three types based on the time period over
which the signal is measured: long-term or aging (over
hours, days or months), medium term or stability (over
second to hours), and short term or noise (second or
less). Usually stability is given in terms of normalized
deviation from target value and measure is the part-per-
million (ppm). A silicon MEMS resonator suitable for high
frequency reference applications should have the frequency
stability approaching 0.1 ppm. Recently fabricated MEMS
oscillator [11] with epi-seal process satisfied this frequency
stability conditions. At the same time it had the power
consumption less than 20mW and it occupied 0.3 mm
3
.
During the past 40 years, multiple fabrication
processes have been mastered that enabled the above
mentioned results to be achieved. Among the most
important processes are encapsulation of the resonant
beam, highly accurate positioning of excitation electrodes
and, finally, ovenization for keeping the resonator at the
exact temperature. As we mentioned earlier, AD processes
can have a significant influence on the short-term stability
through the AD noise. As discussed earlier, the resonant
frequency is proportional to the square root of the inverse
of its mass. The mass of the resonator is of the order of 100
picograms. So physical contamination, equivalent to a
single atomic layer of additional mass deposition, can
change the frequency by hundreds or thousands of ppm.
D. Nanomechanical computers
Recenty, Robert H. Blick et al. [13] proposed a fully
mechanical computer (NMC) based on nanoelectro-
mechanical elements. The main motivation behind
constructing such a computer is threefold: (i) mechanical
elements are more robust to electromagnetic shocks than
current dynamic random access memory based purely on
complementary metal-oxide semiconductor technology
(CMOS), (ii) dissipated power can be an order of
magnitude below CMOS and, (iii) the operating
temperature of such an NMC can be an order of magnitude
above that of conventional CMOS.
Without going into details of NMC design, it is
important to emphasize here that its fundamental element is
a nanoelectromechanical single electron transistor (NEM
SET). In this type of SET [13] the island is movable and it
is situated on the top of a cantilever which can be excited
by an AC source-drain voltage. Recent measurements have
shown that self-excitation can be exploited to generate
mechanical oscillation without any AC excitation. This
means that DC voltage is sufficient to operate the NMC.
Besides its practical value, research of NEMSET is of
great importance for fundamental sciences, since it
incorporates both the field of single electron devices and
the vibrating nanostructures whose dimensions are such
that they can represent the ideal models for demonstration
of the quantum mechanics laws.
III. CONCLUSION
Vibrating micro- and nanostructures are finding an
ever-increasing number of applications in a multitude of
fields: as basic building blocks of the new generation of
sensors with sensitivity of the order of single molecules, as
elements of the new kind of MEMS oscillators with low
power consumption and fabrication process compatible
with CMOS, as central components of NEMSETs etc.
In this paper we presented some of the fundamental
theories upon which the operation of these components is
based. The emphasize was on the fact that in case of
miniature vibrating structures the effects pertinent to their
dimensions, such as Brownian motion, AD processes,
mass, temperature and frequency fluctuations etc. become
significant. The author hopes that the results of this work
can be utilized for proper design and optimization of
various vibrating micro- and nanocomponents.
As it can be expected in such a propulsive field,
several interesting phenomena, such as the quantum effects
in nanovibrating structures, vibrating structures with
nanotubes, multilayer adsorption processes on nano-
structures etc. had to remain out of scope of this paper.
Anyway, research in this field is expected to enable both
better understanding of the nature itself and wise utilization
of its resources.
ACKNOWLEDGEMENT
The author wishes to thank Ms. I. Joki, Mr. M.
Frantlovi and Dr K. Radulovi for their collaboration and
to the Serbian Ministry of Science for their support.
REFERENCES
[1] G. Binning, C. F. Quate, and Ch. Gerber, "Atomic Force
microscope", Phys. Rev. Lett., 1986, vol. 56, pp. 930-933.
[2] N. V. Lavrik, M. J. Sepaniak, and P. G. Datskos, "Cantilever
transducers as a platform for chemical and biological
sensors", Rev. Sci. Instrum., 2004, vol. 45, pp. 2229-2253.
[3] J. L. Arlett et al., BioNEMS : Nanomechanical Systems for
Single-Molecule Biophysics, Lect. Note. Phys., 2007, vol.
711, pp. 241-270.
[4] P. S. Waggoner, and H. G. Craighead, "Micro- and nano-
mechanical sensors for environmental, chemical and
biological detection", Lab on Chip, 2007, vol. 7, pp. 1238-
1255.
[5] K. M. Goeders et al., "Microcantilevers: Sensing Chemical
Interactions via Mechanical Motion", Chem. Rev., 2008, vol.
108, pp. 522-544.
[6] J. E. Sader, "Frequency response of cantilever beams
immersed in viscous fluids with applications to atomic force
microscope", J.Appl.Phys., 1995, vol. 84, pp. 64-76.
[7] C. A. Van Eysden, and J. E. Sader, "Frequency response of
cantilever beams immersed in viscous fluids with application
to the atomic force microscope", J. Appl. Phys., 2007, vol.
101, pp. 044908 1-11.
[8] S. Basak, "Dynamics of oscillating microcantilevers in
viscous fluids", Ph.D. thesis, Purdue University, 2007.
[9] D. M. Karabacak, "Resonant operation of nanoelectro-
mechanical systems in fluidic environments", Ph.D. thesis,
Boston University, 2008.
[10] Z. Djuri et al., "A New Approach to IR Bimaterial Detector
Theory", Infrared Phys. and Techn., 2007, vol. 50, pp. 51-57.
[11] M. A. Hopcroft, "Temperature-stabilized Silicon Resonators
for Frequency Reference", Ph.D.Thesis, Stanford Univ., 2007.
[12] M. L. Roukes, "Mechanical Computation, Redux?", IEDM
Technical Digest, 2004, pp. 539-542.
[13] R. H. Blick, H. Quin, H-S. Kim, and R. Marsland, "A
nanomechanical computer-exploring new avenues of
computing", New J. of Phys., 2007, vol. 9, pp. 241-250.
[14] H. A. C. Tilmans et al., "Micro resonant force gauges", Sens.
Actuators A, 1992, vol. 30, pp. 35-53.
[15] Z. Djuric, "Mechanism of noise sources in microelectro-
mechanical systems", Microelectronics Reliability, 2000, vol.
40, pp. 919-932.
[16] B. Weiss et al., "Modeling of a clamped-clamped beam
vibrating in a fluid for viscosity and density sensing regarding
compressibility", doi:10.1016/j.sna2007.11.029.
[17] H. B. Callen, and T. A. Welton, "Irreversibility and
generalized noise", Phys. Rev., 1951, vol. 83, pp. 34-40.
[18] T. Fukuma et al., "Development of low noise cantilever
deflexion sensor for multienvironment frequency-modulation
atomic force microscopy", Rev. Sci. Instr., 2005, vol. 76,
053704.
[19] D. Ham et al., "A phase noise in oscillators",
www.ush.chiba-u.ac.jp/~ken/Symp2004/PDF/1C3.
[20] T. R. Albrecht et al., "Frequency modulation detection using
high-Q cantilevers for enhanced force microscope
sensitivity", J. Appl. Phys., 1991, vol. 69, pp. 668-673.
[21] Z. Djuri, O. Jaki, and D. Randjelovi, "Adsorption
desorption noise in micro-mechanical resonant structures",
Sens. and Actuators A, 2002, vol. 96, pp. 244-251.
[22] Z. Djuri, I. Joki, M. Frantlovi, and O. Jaki,
"Fluctuations of the number of particles and mass adsorbed on
the sensor surface surrounded by a mixture of an arbitrary
number of gases", Sens. Actuators B, 2007, vol. 127, pp. 625-
631.
[23] Z. Djuri, I. Joki, and M. Frantlovi, "Analysis of transient
adsorption processes using micro/nanocantilever oscillators",
Microelectronic Eng. (2007), doi: 10.1016/j.mee.2007.12.028.
[24] Z. Djuri, and I. Joki, "Thermomechanical noise of
nanooscillators with time-dependent mass", Microelectronic
Eng., 2007, vol. 84, pp. 1639-1642.
[25] N. R. Paul, and M. C. Cross, " Stochastic Dynamics of
Nanoscale Mechanical Oscillators Immersed in a Viscous
Fluid", Phys. Rev. Lett. 2004, vol. 92, pp. 235501 1-4.