Linearity and Sensitivity Issues in Piezoresistive Pressure

Sensors
Jaspreet Singh, M M Nayak1, K Nagachenchaiah
Semiconductor Laboratory(SCL), Dept. of Space, Punjab, India
1
Liquid Propulsion System Centre(LPSC), ISRO, Dept. of Space, Bangalore, India
jaspreet@sclchd.co.in, mmnayak@gmail.com, chen@sclchd.co.in

Abstract
This paper describes the various nonlinearities (NL) encountered in the Si-based Piezoresistive pressure sensors.
The effect of diaphragm thickness and position of the piezoresistors are analyzed taking anisotropy into account.
Also the effect of the oxide/nitride used for isolation between metal and diaphragm is studied from linearity
point of view
Keywords: Pressure sensor, sensitivity, linearity, piezoresistance, wheatstone bridge

1 Introduction 2 Nonlinearities in Piezoresistive

MEMS based Pressure sensors are mechanically
similar to traditional sensors with the exception that
these are Si based and on micrometer scale. The
additional advantages of MEMS based pressure
sensors include batch fabrication, high performance,
small size, low cost, absence of adhesive bonding
layer and easy integration with electronics on single
chip. Pressure sensors have a wide-range of
applications in various fields like automotive industry,
biomedical, space applications and military
pressure sensor
A Piezoresistive pressure sensor consists of a
diaphragm with diffused piezoresistors in Wheatstone
bridge configuration (Fig.–1). The diaphragm
converts pressure into mechanical stress, the
piezoresistors converts this stress into resistance
change and finally the resistance change is converted
the into output voltage.
APPLIED
PRESSURE
(INPUT)

applications. These pressure sensors are available in

wide operating range covering from fractions of psi to
15,000 psi. Si
DIAPHRAGM
STRESS PIEZO
RESISTORS
RESISTANCE WHEAT
STONE VOLTAGE
BRIDGE OUTPUT

A lot of research has been carried out on

micromachined piezoresistive pressure sensors in the
recent years [1-4]. For high performance demands, the
SUPPLY
sensitivity and linearity must be improved. In order to VOLTAGE

increase the sensitivity, the diaphragm thickness

should be thin. In present commercial Piezoresistive
pressure sensors, Si diaphragms with less than 20 m
thickness are common[5]. Generally thin diaphragms
are prone to large deflections and nonlinear effects. It
is therefore necessary to optimize the diaphragm
thickness with respect to rigidity and strength.
Nonlinear higher order piezoresistance coefficients
add-up the linearity error further. Proper selection of
piezoresistors i.e. orientation, shape, location, doping
concentration, dose etc. is essential. Finally the
conversion of small resistance change to voltage
output is another additive nonlinearity.
Figure – 1: Principle of Piezoresistive pressure sensor
These subsystems have to be considered and
optimized in order to realize a pressure sensor with
high sensitivity and good linearity. Nonlinearity of a
transducer can be defined as the maximum deviation
of the calibration curve from specified best fit straight
line. Mathematically, overall non-linearity for
Piezoresistive pressure transducer can be given as
NL = NL2 p ,d + NL2 d ,r + NL2 p ,r (1)

This paper reviews the types of nonlinearities in the
context of Piezoresistive pressure sensors, giving
basic relationships between pressure, stress,
deflection, and resistance change and voltage output.
The results of numerical simulations of square
diaphragm for optimum load-deflection are presented.
where NLp,d is the nonlinearity between pressure-
deflection (structural nonlinearity), NLd,r is the
nonlinearity between deflection-resistance
(piezoresistive nonlinearity) and NLp,r is the non-
linearity due to difference in the sensitivities to
pressure among resistors (bridge nonlinearity). In
what follows, these nonlinearities are discussed in
some detail.

1
3 Structural Nonlinearity hE  ∂ 2 w ∂2w 
σ xx =  + υ 
Consider a thin silicon plate subjected to an applied 2(1 − υ 2 )  ∂x 2 ∂y 2 
pressure p resulting in lateral bending. The governing
differential equation can be written as [6-7] hE  ∂ 2 w ∂2w 
σ yy =  + υ 
2(1 − υ 2 )  ∂y 2 ∂x 2 
(5)

∂ 2M x ∂ 2 M xy ∂ 2 M y  ∂ 2w 
−2 + = −p τ xy = hG 
 ∂x∂y 
(2)
∂x 2 ∂x∂y ∂y 2

The above description is just a summary of general

The above equation (2) can be written in terms of practice used for calculating stresses and deflection of
load-deflection form as a square plate clamped at the edges. The detailed
analysis is based on small deflection theory. It
assumes that the stress distribution is a result of pure
∂ 4w ∂ 2w ∂ 4w p bending i.e. neutral plane of the diaphragm is not
+ 2 + = (3) stretched. This assumption requires that the deflection
∂x 4 ∂x 2 ∂y 2 ∂y 4 D of the diaphragm be small when compared with its
thickness. For thin diaphragms, generally encountered
in pressure sensors, this analysis is not sufficient. In
where w(x,y) is the deflection which can be found by case of thin diaphragm, if deflection is not small,
solving (3) with proper boundary conditions. D refers neutral plane of the diaphragm will stretch like a
to flexural rigidity assuming the constant plate balloon. It is called “balloon effect” [13]. The stress
thickness h. Various computer codes [8-9] are caused by the stretch of neutral plane has to be
available to solve such equations with proper considered in that case.
boundary conditions. The bending strains at the The factors that are contributing to structural
surface can be written as: nonlinearity can be grouped as
(i) Geometric nonlinearity
(ii) Material nonlinearity
(iii) Contact nonlinearity
Geometric nonlinearity occurs when there are large
displacements under specified loading conditions. In
these cases the small deflection theory is not sufficient
to give the deformation behaviour of the plate in full
working range. Now strain displacement relations are
no longer linear:
2
∂u 1  ∂w 
ε xx = +  
Figure – 2: Plate subjected to uniform pressure load ∂x 2  ∂x 
2
∂v 1  ∂w 
ε yy = +  
h ∂2w ∂y 2  ∂y 
ε xx =−
2 ∂x 2  ∂u ∂v   ∂w ∂w 
ε xy =  +  +  . 
h ∂2w  ∂y ∂x   ∂x ∂y 
ε yy =− (4) (6)
2 ∂y 2
where u, v and w are displacements in x, y and z
∂ 2w directions respectively. In this case the stress in
ε xy = −h diaphragm consists of two parts: the first caused by
∂x∂y bending of the diaphragm and second stress caused by
Using constitutive equations, the stresses can be found the stretch of neutral plane. Mathematically,
as:
σ = σ bending + σ stretch
(7)

2
Here the load (pressure) is shared by the stretch action deposition/growth, implantation and anisotropic
also, so bending stress will reduce as compare to the etching etc. contribute to the nonlinearity. Some of
value calculated by small deflection theory. these stresses can be relieved during fabrication
process itself like drive-in, annealing etc. However
complete removal of the stress is not always possible
6
since these are not completely known- qualitatively
4 and quantitatively. Thus the total stress in diaphragm
will be:
LinearityError(%FSO)

0 σ = σ bending ± σ residual
0 0.2 0.4
-2
Pressure(bar)
-4
-6
-8
-10
Figure – 3: Linearity error (structural) with and
without considering large deflection effect
Figure -3 shows simulation results of linearity error
with and without accounting the effect of large
deflections for a square diaphragm of width 1054 µm
and 10 µm thickness used for 1 bar pressure sensor.
The analysis is done using ANSYS. As can be seen,
the effect of large deflections is quiet high in this case.
The curve is now no longer symmetrical with respect
to central point. So the diaphragm is redesigned with
optimum dimensions for the same pressure range.
Figure-4 shows the results of linearity error with
0.6 0.8 1 (8)
It is found [10] that tensile residual stresses increase
the bending stiffness (higher stress leads to higher
Small deflection
theory stiffness) of the plate while compressive residual
Large Deflections stresses reduce the stiffness and could eventually lead
effect
to buckling. This effect was investigated by
simulating the diaphragm with grown oxide (having
compressive stress) and with both oxide and CVD
nitride (CVD Nitride has tensile stress) [11]. It is seen
from figure – 5 that the effect of 0.1 micron grown
oxide is negligible as compare to bare Si diaphragm
(without oxide). The output increases due to
compressive stress and also the structural nonlinearity
by about 0.02% FSO. Similarly the effect of
Oxide/Nitride layer on the diaphragm is shown in
figure – 6. It shows the degradation of both linearity
as well as output as compare to only oxide layer.
0.4

modified dimensions of 16 m thickness: Without oxide

with 1500A Oxide
0.3

0.2
6

16micron 0.1
4
Linearity error(%FSO)

10micron
0
Linearity error(%FSO)

0 0.2 0.4 presure(bar) 0.6 0.8 1

2
-0.1

0
-0.2
0 0.2 0.4 0.6 0.8 1

-2 -0.3
Pressure(bar)
-0.4
-4

-0.5

-6

-8 Figure – 5: Linearity (structural) error with and

without oxide
Figure – 4: Linearity (structural) error for two
Contact Non-linearity arises due to change in
different thicknesses of the diaphragm. Lesser
boundary conditions. Since edges are assumed to be
thickness more prone to NL
built-in, in case of thin diaphragm this nonlinearity is
The reduction in output signal can be suitably taken negligible compared to the other two described
care of in the electronic signal processing subject to earlier.
meeting the sensitivity requirements. It is also seen
[15] by applying the pressure on the back side of the
diaphragm the geometric nonlinearity is affected
considering large deflections since the sign of b and
s are opposite.

Material non-linearity should be ideally zero in case

of single crystal Si diaphragms as the stress-strain
relation is linear up to fracture point. However,
residual stresses of the fabrication process like

3
 3.0

400/800 oxide nitride 6

with 1500A oxide
2.0

NL(% FSO )
1.0

2
Non-Linearity(%FSO)

0.0
0 0.2 0.4 pressure(bar) 0.6 0.8 1
0

0
-75
-50
-25

25
50
75
100
125
150
175
200
-200
-175
-150
-125
-100
-1.0
-2
-2.0
-4
-3.0
Stres(MPa)

-4.0
Longitudinal Stress Transverse Stress

Figure – 6: Linearity error with and without Figure – 7: Nonlinearity of p-type piezoresistors for
oxide/nitride combination <110> stress (doping level: 2x1018/cm3)

4 Piezoresistive Nonlinearity Based on this, one of the techniques adopted for

For a diffused piezoresistor subjected to parallel and
perpendicular stress components, the resistance
change is given by
∆R ∆ρ
≈ = ∏l σ l + ∏t σ t
R ρ
where l and t are piezoresistive coefficients parallel
and perpendicular to the resistor length. This relation
assumes that stress levels are relatively small and
hence Piezoresistive coefficients of Silicon are
independent of stress i.e. when stresses are linear with
applied pressure, resistance change will be linear with
reducing this nonlinearity is by using only transverse
piezoresistors instead of both transverse and
longitudinal piezoresistors. The effect Piezoresistive
non-linearity is seen by diffusing suitable resistors on
the diaphragm in full Wheatstone configuration. It
was observed that “structural” non-linearity is
(9)
partially compensated by the non-linearity of the
Piezoresistive effect. Figure-8 shows how structural
linearity error changes from 0.37% to 0.23%FSO for
16 m diaphragm with piezoresistors.
0.4
structural NL
structural+Piezoresitive
NL
0.3
0.2

stress. But, in actual practice extra amount of 0.1

nonlinearity is observed. This nonlinearity is due to

Non-Linearity(%FSO)

the dependence of piezoresistive coefficients on the 0 0.2 0.4 pressure(bar) 0.6 0.8 1

-0.1
stress. However, investigation of the dependence of
piezoresistive coefficient on stress is quite involved as -0.2

there are many components of stress tensor and the -0.3

measurement of higher order effects requires very -0.4

high accuracy. The magnitude of this nonlinearity is

-0.5
proportional to the stress value. It has been
investigated [12] that nonlinear Piezoresistive
coefficients up to third order can play major role in Figure – 8: Structural NL and Overall NL for 0-1bar
certain crystallographic directions. Up to first order pressure sensor
11, 12 and 44 can give l and t for any arbitrary
direction in the crystal. These three coefficients 5 Bridge Nonlinearity
further are functions of doping concentration and
temperature. But for second order, nine more such When an external pressure is applied, the diaphragm
piezoresistance components are needed to calculate l is stressed and the longitudinal and transverse
and t. The observation of Matsuda et al [12], for p- resistors undergo different changes in resistances due
type resistors oriented in <110> orientations with a to the average stresses being different in each resistor.
doping level of 2x1018/cm3, the dependence of The expression for the resistance change is given in
nonlinearity on stress is shown in figure - 7. As can be equation (9). Generally, all the four resistors are
seen, the nonlinearity due to piezoresistive effect for connected in Wheatstone bridge – either full active or
longitudinal resistor is positive for both tensile and partial. Considering full Wheatstone bridge
compressive bending stresses whereas for transverse configuration the output voltage can be calculated as:
resistor, NL is negative for compressive and positive Vo (∆R R )l − (∆R R )t
for tensile stresses. It is found that third order = (10)
polynomial approximation gives fairly good match Vs 2 + (∆R R )l + (∆R R )t
[14].

4
Ideally, in a linear voltage output bridge the output is
proportional to the deflection of the membrane and
hence to the applied pressure.
The denominator of the above expression (10)
introduces nonlinearity which can be eliminated by
designing the resistors such that
 ∆R   ∆R 
  =  (11)
 R l  R t
i.e. sensitivity among the piezoresistors should be
same. If the above condition is met then the
expression for the output voltage ratio is equal to the
The piezoresistors have to be placed properly on the
diaphragm. For p-type resistors aligned in <110> on
(100) Si wafer, piezoresistive coefficients ( l and t)
are almost equal in magnitude but opposite in sign,
the bridge configuration allows maximizing the
sensitivity of the output signal.
6 Experimental Results
The simulation results were experimentally verified
by fabricating the device. Figure – 9 shows the plot of
observed as well as calculated nonlinearity for 0-1bar
pressure range.
0.4
Simulated
Experimental
0.3

fractional change in piezoresistance of the resistors. 0.2

This nonlinearity can be understood by considering

0.1
the linear but different resistance change with pressure

Linearity Error(%FSO)
in longitudinal and transverse piezoresistors. The 0
0 0.2 0.4 0.6 0.8 1

change in resistance is given by -0.1

pressure(bar)

 ∆R 
-0.2

  = αp -0.3

 R l -0.4
(12)
 ∆R 
 = − βp
-0.5


 R t
Figure – 9: Experimental and simulated nonlinearity
where , denotes the sensitivities of the resistors
respectively. Substituting (11) in (9), and assuming The effect of increase in the thickness and removing
offset voltage of the bridge is zero, the output voltage nitride from the diaphragm is given below in table– 1:
(V0) is
(α + β ) p 10 m 16 m Improveme
Vo = Vs (13) diaphragm diaphragm nt with
2 + (α − β ) p
thickness with 1000Å modified
The end point nonlinearity at a specific test pressure pi with 500Å oxide design
can be given as oxide and
800Å nitride
Vo ( p m )
Vo ( pi ) − pi Nonlinearit 2.5 0.40 6.25 times
pm (14)
NLi (% FSO) = x100 y (%FSO)
Vo ( p m )
Hysterisis 0.7 0.3 2.33 times
Substituting (12) in (13) and assuming maximum (%FSO)
nonlinearity for the whole operation range is at pi =
pm/2, i.e. half of maximum applied pressure (pm), the Sensitivity 20 15 -0.75 times
NL will be (mV/V/bar)

p m  (α − β )  Table 1: Nonlinearity and hyteresis improvement

NL(%FSO) =   x100 (15)
2  4 + (α − β ) p m 
It is clear from (15) that as the difference between
and increases the linearity error will increase.
Further, in above equation, it was assumed that the
resistance sensitivity varies linearly with pressure, but
in actual practice it is not the case. So the difference
between resistor sensitivities can still be higher and
hence the linearity error. Generally, this nonlinearity
component has least effect if location of the
piezoresistors is well optimized.
As can be seen form table – 1, improvement of the
order of 6.25 times and 2.33 times in nonlinearity and
hysteresis respectively is achieved in the modified
design. The loss of sensitivity can be attributed due to
increase in the thickness which is only 0.75 times less
than with 10 m thickness.
7 Conclusions
Linearity and sensitivity are the two performance
parameters of the pressure sensor, which are traded
off in the realization of the sensor. An analysis of the

5
various issues involved in the performance [6] Timoshenko S. P. and Woinowsky-Krieger,
optimisation is presented in this paper. It is Theory of plates and shells, 2nd edition New York
analytically as well as experimentally found that using McGraw-Hill, 1970.
16µm thick diaphragm in place of 10µm shows good
[7] Elgamel, H.E., “Closed-form expressions for the
linearity response of the order of 6.25 times with
relationships between stress, diaphragm deflection,
tolerable loss in sensitivity which is 0.75 times w.r.t
and resistance change with pressure in silicon
10 m thickness. Depending on the causes of
piezoresistive pressure sensors” Sensors and
nonlinearities, the following approaches are suggested
Actuators A, 50 (1995) 17-22.
to reduce/eliminate the same:
[8] MATHEMATICA help manuals
(a) Using optimum thickness of the diaphragm.
[9] MATLAB, PDE Tool box
(b) For isolation purposes between metal and
diaphragm, avoiding use of nitride (It is [10] Senturia S.D. Microsystem Design, Kluwer
always better to use only oxide). If oxide Academic Publishers, London,2003
nitride stack is used, thicknesses of these
[11] Hu S M “Stress related problems in Silicon
should be properly chosen to have minimum
technology” J. Appl. Phy. 70(6) ppR53-R73, 1991
residual stress effect.
[12] K Matsuda et al. “Nonlinearity of piezoresistance
(c) Geometric non-linearity can also be well
taken care by making bending and stretch of effects in p- and n-type silicon” Sens. and Act. A 21-
opposite nature. 23 pp 45-48, 1990
[13] Lin L, Chu, Lu Y W “Simulation program for the
(d) Piezoresistor design to be such that ∆R shall sensitivity and linearity of piezoresistive pressure
R sensors” Jour. Of microelectromechanical sens. Vol.
have linear response with load/stress. This
8 no. 4 pp. 514-522, 1999
can be ensured by proper placement of the
resistors. [14] Kazuji Yamda et al. “Nonlinearity of the
piezoresistance effect of p-type Silicon diffused
8 Acknowledgement layers”. IEEE Trans. On Elec. Devices, vol. ED-29
No.1 pp.71-77, 1982.
The work is carried out at Semiconductor Laboratory.
The authors wish to thank chairman ISRO and the [15] Minhang Bao, Principles of MEMS Devices,
Directors VSSC and LPSC for their encouragement. Elsveir Publishers, Ed.2005
Also, the fruitful discussions and cooperation of their
colleagues at SCL is gratefully acknowledged.

9 References
[1] Samuel K Clark and Kensall D Wise “Pressure
sensitivity in anisotropically etched thin diaphragm
pressure sensors” IEEE Tran. Of Elec. Devices, vol.
ED-26, no. 12, pp. 1887-1895, 1979
[2] H.L.Chau and K.D.Wise “Scaling limits in batch
fabricated silicon pressure sensors” Sens. Actuators,
vol. 10, pp. 303-320, 1986
[3] Zhadko I.P., Babichev G.G. “Silicon pressure
transducer with differential sensitive element based on
transverse electromotive force” Sens. and Act. A 90
pp.89-95, 2001
[4] Lynn F Fuller, “Bulk micromachined pressure
sensor”University/Government/Industry,
Microelectronics Symposium, Proceedings of
the15thBiennial, pp.317-320, 2003.

[5] Zhang Y et al “Design, fabrication and

characterization of novel piezoresistive pressure
microsensor for TPMS” IEEE trans. of Elec. Devices,
2006 pp. 443-446