Sensors and Actuators A 119 (2005) 476–488

Optimal energy density piezoelectric bending actuators

R.J. Wood∗ , E. Steltz, R.S. Fearing
Department of EECS, University of California at Berkeley, 211 Cory Hall 1772, Berkeley, CA 94720, USA

Received 14 June 2004; received in revised form 6 October 2004; accepted 17 October 2004
Available online 8 December 2004

Abstract
The design and analysis of piezoelectric actuators is rarely optimized for low mass applications. However, emerging technologies such
as micro air vehicles, and microrobotics in general, demand high force, high displacement, low mass actuators. Utilization of generic
piezoceramics and high performance composite materials coupled with intelligent use of geometry and novel driving techniques yields low
cost, rapidly prototyped, ultra-high energy density bending actuators for use in such applications. The design is based upon a laminate
plate theory model for a stacked multimorph cantilever actuator, encompassing all possible layups, layer anisotropies, internal and external
excitations, and intrinsic and extrinsic geometries. Using these principles, we have fabricated 12 mg PZT bimorph actuators with greater than
2 J kg−1 energy density. This gives a performance increase of an order of magnitude or greater compared to existing commercially available
piezoelectric bending actuators.
© 2004 Elsevier B.V. All rights reserved.

Keywords: Piezoelectric actuators; Microactuators; Bimorph; Composites

1. Introduction morph actuator and compares the results to existing bending

actuators.
This work addresses a number of key points in actuator de- One novelty of the approaches presented here is the mod-
sign, specifically for high energy density applications, which eling of the actuator performance, both of the actuator output
are unique among the design of transducers. The overall goal and of the interactions between internal and external excita-
of this paper is to describe methods for creating bending ac- tions. There have been numerous models presented for the
tuators which result in each infinitesimal electro-active ele- mechanics of rectangular piezoelectric transducers. DeVoe
ment having a strain close to its ultimate strain for a given and Pisano [4] presented a model for MEMs cantilever actu-
set of known internal and external excitations. In total, three ators considering multiple passive layers. Smits and Ballato
coupled techniques are described here for increasing the en- [13], Smits and Choi [14] and Weinberg [18] described in
ergy density of clamped-free piezoelectric bending actua- detail a one-dimensional analysis of piezoelectric bending
tors: geometry, smart material choices, and optimal high field actuator performance using energy methods. More specific
driving techniques. Each individually gives an improvement to this work, Wang and Cross [16] modeled a bimorph with
over existing competitors, however when coupled together a central passive layer and the effects thereof. For micro-
the result is a factor of 10 or greater energy density im- robotics applications, Sitti et al. [12] described the design
provement. As a verification of the model and fabrication of millimeter scale bending actuators. For greater general-
process, this work develops and tests a millimeter-scale bi- ity, laminate plate theory is used in this work to describe
the interaction between the external and internal forces and
moments with the layer stresses and strains. This not only
∗ Corresponding author. Tel.: +1 510 6435796; fax: +1 510 6427644. predicts the displacement and blocked force of the actu-
E-mail addresses: [email protected] (R.J. Wood); ator, but additionally the strains in each layer for failure
[email protected] (E. Steltz); analysis as well. Although the discussion in this paper is
[email protected] (R.S. Fearing)

0924-4247/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.sna.2004.10.024
 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488 477

Fig. 1. Bimorph drawing with pertinent dimension descriptions.

concentrated on piezoelectric actuation, this model is easily as the area under the force–displacement curve:
adapted to thermally excited and other electro-active actu-
ators as well. The use of actively or passively (elastically) Um = 21 Fb δmax (1)
anisotropic constituent layers can also yield interesting tai-
where Fb and δmax are the peak-to-peak blocked force and
lored kinematics (bending-twisting and extension-twisting
unloaded maximum tip displacement respectively for a given
coupling for example) and is also encompassed within this
field. This paper assumes that the actuators are driven quasi-
model.
static (f  f0 ), thus the power is a linear function of the
There are a number of existing and conceptual applica-
drive frequency. However for such actuators driving resonant
tions for such a technology, for example as the flight mus-
systems ([2,20,21]) the power able to be delivered to a given
cles for a flapping wing micro air vehicle [21], control sur-
load is a function of internal dissipations. The dissipation
face actuators for indoor slow fliers [6,9], motors for mi-
effects on the power delivery capabilities for resonant systems
cro legged robots [5], drivers for haptic display devices
are quantified in Eq. (2).
[11], etc.

1
2πf0 Fb δQl for Qa  Ql
P = 81 (2)
16 2πf0 Fb δQ for Qa ≈ Ql
1.1. Design preliminaries
In Eq. (2) f0 is the total system resonant frequency and the
Fig. 1 shows a drawing of the proposed bimorph actuator mechanical Q is a function
√ of the lumped system stiffness,
for design purposes. Note that the width is tapered along the mass, and loss: Q = km/b. This paper does not address
length; the actuator geometry will be discussed in Section 2.1. actuator and load dynamics and losses and thus the energy
Also, the elastic (passive) material is drawn as one layer. In density will be the performance merit for the remainder of
reality, however, it could be composed of a number of layers this discussion.
in arbitrary orientations as will be discussed in Section 2. The Table 1 lists the specifications of current commercially
effects of an extension will be described in Section 2.2. For available actuators. Note that the magnitudes of the applied
performance comparisons, the mechanical energy is defined fields in Table 1 for the first two actuators are small compared

Table 1
Commercially available clamped-free cantilever piezoelectric bending actuator specifications
Actuator δmax a (␮m) Fb a (mN) m (mg) DU (J kg−1 ) Fieldb (V␮m−1 ) Piezo materialc
T219-H4CL-103Xd 610 160 320 0.153 0.25 5H
QP21Be 790 460 2800 0.065 0.50 5A
TH-8Rf 1900 111 1780 0.059 1.75 5H
Optimized bimorphg 520 123 12 2.730 2.36 5H
Maximum strain energy density for bulk free plateh 4.0 2.5 5H
a Peak-to-peak.
b Maximum drive field.
c Either PZT-5H or PZT-5A.
d Piezo Systems (http://www.piezo.com).
e Mide QuickPack actuators (http://www.mide.com/quickpack/qp pricelist.html).
f THin layer UNimorph DrivER and sensor from Face Thunder (from empirical measurements and http://www.face-int.com/thunder/thunder.htm).
g Strain-optimized bimorph micro-actuators described in this work.
h For d31 actuation.
478 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488

to the field applied to the bimorphs in question (>2 V␮m−1

as will be discussed in Section 4). There are four key factors
that can limit the magnitude of field applied to piezoelec-
tric actuators: mechanical failure (fracture), electrical failure
(dielectric breakdown), depolarization, or saturation of the
piezoelectric effect. Bimorphs connected in series or paral-
lel to the drive source are limited by depolarization (as is
the case with the first two actuators in Table 1). Unimorphs
are not subject to depoling so long as the field is unipolar
in the poling direction, or a small magnitude bipolar field.
The THUNDER actuators in Table 1 are unimorphs and the
field limit listed is based upon commercial specifications. Fig. 2. Actuator layup for arbitrary lamina materials and ply angles.
The actuators described here are limited by breakdown and
mechanical failure; little saturation has been observed be-
fore either electrical or mechanical failure. It is important to given by the following:
note that it is not only the driving method that allows these
1
actuators to be driven at such high fields. The intrinsic and ex- 1 = σ1 + d31 E3 p + α1 T (4)
trinsic geometry modifications give the capability to run the E
actuator at fields which would fracture traditional rectangu- where σ is an applied stress, E3 the electric field, α the co-
lar bimorphs. All peak field data for commercially available efficient of thermal expansion (CTE), and T the change in
actuators shown in Table 1 are directly quoted from the man- temperature, which for this application is the change from
ufacturer. the cure temperature to room temperature (curing details are
given in Section 3). The p term in (4) is a placeholder which
is defined as follows:
2. Laminate plate theory for the design of multilayer 

1 field parallel to piezoelectric poling
bending actuators
p = −1 field antiparallel to piezoelectric poling (5)


Throughout the following discussion a number of assump- 0 else
tions are made. First, the piezoelectric materials are assumed
to be transversely isotropic, that is d31 = d32 and there are Thus, this model can be applied to any combination of piezo-
no piezoelectrically induced shearing forces, d36 = 0 [7]. In electric and passive plates. Note that for the case of an
general this is only true for the case of polycrystalline piezo- anisotropic composite material, the strains in (4) are along
electric materials; with single crystal materials the piezoelec- the fiber direction, as is defined in Fig. 3. More generally, the
tric constant will vary with the orientation relative to the crys- in-plain strains assume the following form:
tal directions. However, if the crystal is cut properly so that the      
1 S11 S12 0 σ1
crystal planes are aligned to the actuator geometry, then the      
material is piezoelectrically orthotropic and this assumption  2  =  S12 S22 0   σ2 
holds for the single crystal case as well. Second, there is no ex- γ12 n 0 0 S66 n τ12 n
ternal axial loading, only transverse loading which is applied    
d31 α1
at the distal end of the actuator. Third, the bonding between   n  
each layer is assumed to be perfect, that is, there are no shear +  d32  E3 +  α2  T (6)
strains between layers. Fourthly, for thermal calculations, 0 n 0 n
there is no gradient in temperature through the thickness of
any lamina. Also, electrostriction and higher order effects are The [Sij ]n terms are the compliances of the nth layer. Solving
ignored. The change in effective field due to electromechani- (6) for the stresses in the piezo layer yields the following:
cal coupling (as in [15]) is also ignored. Finally, when the ac-      
σ1 Q11 Q12 0 1
tuator width is much greater than the thickness, a plane strain      
state is incurred where y ≡ 0 [18]. This causes the Young’s  σ2  =  Q12 Q22 0   2 
modulus and piezoelectric properties to be modified as τ12 n 0 0 Q66 n γ12 n
follows:     
d31 α1
    
Ei → Ei (1 − νi2 )−1 , d31 → d31 (1 + νi ) (3) −  d32  E3n −  α2  T  (7)
0 n
0 n
Fig. 2 shows the cross-section of a laminate consisting of an
arbitrary lamina layup. This will be used to define the lamina In Eqs. (6) and (7), the [Qij ]n terms are the plane strain
geometry throughout this section. The strain in any layer is modified material constants of the lamina as given in Table 2.
 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488 479

Fig. 3. Lamina axes diagram for (a) plate aligned to global axes and (b) arbitrary orientation.

To generalize this to arbitrary lamina orientations, the follow- Next, the actuator properties are determined as a function
ing notation is used: of the ply layup using laminate plate theory. First, the rela-
      tionship between the midplane strains and curvatures and the
σx Q̄11 Q̄12 Q̄16 x forces and moments is given by:
     
 σy  =  Q̄12 Q̄22 Q̄26   y      
τxy Q̄16 Q̄26 Q̄66 n γxy n N Aij Bij 0
n = (13)
     M Bij Dij κ
d31 ᾱx
    
−  d32  E3n −  ᾱy  T  (8) In Eq. (13) the A, B, and D terms are given as follows:
0 n ᾱxy n 
Aij = [Q̄ij ]n (zn − zn−1 ),
n
Where the new [Q̄ij ]n is the adjusted stiffness matrix whose
elements have the following properties: 1
Bij = [Q̄ij ]n (z2n − z2n−1 ),
2 n
[Q̄ij ] = [T ]−1 [Qij ][T ]−T (9)
1
and the transformed CTE terms [ᾱi ] are as follows: Dij = [Q̄ij ]n (z3n − z3n−1 ) (14)
3 n
[ᾱ] = [T ]−1 [α] (10)
In Eq. (14), the term zn is the directed height of the nth lamina
where the transformation matrix [T] is defined in (11). with respect to the mid plane as is shown in Fig. 2. In Eq. (13),
 2  the total forces and moments per unit width [ N M ]T can be
m n2 2mn split up into three terms: the externally applied moments and
 
[T ] =  n2 m2 −2mn  (11) the internal forces and moments from the piezoelectric effect
−mn mn m − n 2 2 and thermal expansion all per unit width.
       
In (11), the terms m and n are cos(θ) and sin(θ) respectively N Next Np Nt
= + + (15)
where θ is the angle between the global axes and the lamina M Mext Mp Mt
fiber direction (see Fig. 3(b)). Now the forces and moments
(per unit width) are given as a function of the ply stresses: The piezoelectric forces and moments are defined as follows:
  zn   zn
p
[Ni ] = [σi ]n dz, [Ni (E3 )] = [Q̄ij ]n d3j E3 p dz,
n zn−1 n zn−1

 zn  zn
[Mi ] = [σi ]n z dz (12) [Mi (E3 )]p = [Q̄ij ]n d3j E3 z p dz (16)
n zn−1 n zn−1

Table 2
Design parameters for actuator materials
Parameters UHM CFa S2Glassa Steel Si PZT-5H PZN-PT Units
E1 350 60 193 190 62 15 GPa
E2 7 7 193 190 62 15 GPa
ρ 1500 1600 7800 2300 7800 8300 kg m−3
d31 – – – – −320 −950 pmV−1
σu,1 b 840 1400 900c 120 200d 40d MPa
a Cured.
b Ultimate stress.
c Yield stress.
d Estimated from empirical observations.
480 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488

Similarly, the thermal expansion forces and moments are: 2.1. Actuator geometry
  zn
[Ni ]t = [Q̄ij ]n [ᾱj ]n T dz, For the case of transverse external loading, significant
n zn−1 tensile stresses will be developed in the outer layers. For
 zn the case of a clamped-free cantilever, the moment per unit
[Mi ]t = [Q̄ij ]n [ᾱj ]n Tz dz (17) width generated in the beam is Mx (x) = −F (l − x)/w. For
n zn−1 a constant cross-section, this moment is proportional to the
stress at a given point x, thus the stresses will be max-
Solving Eq. (13) for the midplane strains and curvatures
imum at the proximal end of the actuator. If the cross-
yields the following:
section varies along x, the stress profile can be controlled
   −1       and large stresses can be eliminated. Varying the thickness
0 Aij Bij Next Np Nt
= + + of the piezoelectric material is not practical given the thick-
κ Bij Dij Mext Mp Mt ness of the commercially available PZT plates used (127 ␮m);
(18) however, controlling the width is relatively simple. To ex-
plore this further, consider Eq. (21) and expand the external
Finally, the free displacement and blocked force of the ac- moment.
tuator are found as a function of the applied fields ([E3 ]n )
d2 δ(x) C44 F (l − x)
and the external loading. Since axial strains do not con- = P(E3 ) − (22)
tribute to lateral displacement of the distal end of the dx 2 w(x)
cantilever, the only quantity of interest from Eq. (18) is Thus the free deflection (with F ≡ 0) does not vary with a
κx . First note that the curvature κx is related to the dis- change in the width profile. Next it is necessary to deter-
placement as d2 δ(x)/dx2 = κx where δ(x) is the displace- mine w(x) explicitly for each profile for use in Eq. (22).
ment of the actuator at any point along the x-axis and For the case of a trapezoidal profile, this is given by the
define: following:
 −1
Aij Bij  
C= 2(1 − wr )
(19) w(x) = wnom x + wr (23)
Bij Dij l

Next the external forces and moments are included into For the above equation, wnom is the nominal width (the width
Eq. (18). Note that for a clamped-free cantilever with no at x = l/2) which is the same for all trapezoidal width profiles
external axial forces and an external moment about the y- (to keep the platform area and thus the mass constant for
axis, [ Next Mext ]T = [ 0 0 0 Mx (x) 0 0 ]T . For convenience, collateral comparisons) and the width ratio, wr = w0 /wnom .
define the following: Example width profiles are shown in Fig. 5. In Fig. 5 w0 and
wl are the width at the proximal and distal ends respectively.
P(E3 ) = C41 Nxp (E3 ) + C42 Nyp (E3 ) + C44 Mxp (E3 ) It is trivial to see that for a thin long clamped-free beam,
+ C45 Myp (E3 ) (20) point loaded at the distal end, the strain at the proximal end
is inversely proportional to the width ratio. To illustrate this
where Cij is the (i, j)th element of C (i, j ∈ {1 : 6}). Thus point more concisely the normalized strain profile along the
it can be seen that the curvature is related to the internal length of the actuator is plotted for a few width ratios in
and external moments as follows (ignoring the static thermal Fig. 6(a). Thus it is clear that altering the width can lower
forces and moments): peak stresses, increasing the load to fracture; now it will be
shown that for a fixed actuator area, varying the width can
d2 δ(x) increase or decrease the blocked force. Again the curvature
= P(E3 ) + C44 Mx (x) (21)
dx2 is given by:
A functional diagram of the actuator with respect to the ex-  
d2 δ(x) C44 F l−x
ternal parameters in Eq. (21) is shown in Fig. 4. = P(E3 ) − (24)
dx2 wnom (2(1 − wr )/ l)x + wr

Fig. 4. Actuator diagram with respect to external parameters. Fig. 5. Three representative width profiles.
 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488 481

Fig. 6. Strain profiles for a few width ratios (a) and width factor as a function of width ratio (b) with experimental data (N = 3) for two geometries.

Integrating this twice noting the strict boundary conditions moment as seen by the active material, the strain is more
yields the displacement at the distal end of the actuator: uniformly distributed along the length. Because of this, large
differences in stress between different sections are decreased
P(E3 )l2 C44 Fl3
δ(l) = − and each infinitesimal section of the piezoelectric material
2 wnom can be driven closer to the fracture strain. Thus the extension
 
(wr − 2)2 ln ((2 − wr ) /wr ) − 6 + 10wr − 4w2r does not add mechanical energy to the system, but instead
× allows all parts of the actuator to contribute more uniformly
8(1 − wr )3
to the work. To examine the effects of the extension, first
(25) the external moment term, Mx (x) from Eq. (21) needs to be
Now when δ(l) = 0, the blocked force, Fb , is defined as: determined.
   
3P(E3 )wnom F 1
Fb = W(wr ) (26) = Fext (29)
2C44 l Mext −lext
where W is the width factor is given by the following: Now it is clear that there will be a superposition of a pure
W(wr ) = moment and a force generated by the point load. Thus the
moment per unit width is given by the following:
8(1 − wr )3
(27) Fext ((l + lext ) − x)
3(wr − 2)2 ln((2 − wr )/wr ) − 18 + 30wr − 12w2r Mx (x) = − (30)
wnom
The width factor is plotted in Fig. 6(b), showing the variation
where now Fext = F applied at the tip of the extension. Note
with the width ratio. Note that for wr = 1, W = 1, yield-
that in the above equation, the width is set to be constant
ing the same blocked force prediction as for the rectangular
and equal to wnom . Next, the curvature in Eq. (21) is split
case. Substituting the blocked force and free displacement
into two terms, one from the internal piezoelectric moment
into Eq. (1) will yield the mechanical energy:
(d2 δp (x)/dx2 ) and one from the externally applied moment
3P(E3 )2 wnom l (d2 δf (x)/dx2 ). At x = l, the displacement at the interface
Um = W(wr ) (28)
8C44 between the piezo and extension is found by integrating twice
these two curvatures:
Note that the term wnom l represents the area of the actuator,
and that the energy is linear with the area. Thus the greatest P(E3 )l2
δp (l) = (31)
energy and energy densities are obtained from a width ratio 2
of 2, which represents a triangular actuator.

2.2. Rigid extension

Another method of improving the energy density of a can-

tilever bending actuator is to add an extension to the distal
end. The concept of a rigid extension was introduced by Cam-
polo et al. [3] and is shown in Fig. 7. This extension acts as
a lever which converts the force on the tip to a force and
moment at the interface between the piezo and the extension
as in Eq. (29). By transforming the point load to a force and Fig. 7. Exploded image of actuator with extension.
482 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488

yields the blocked force as follows:

3P(E3 )wnom (1 + 2lr )
Fb,ext = (36)
2C44 l (1 + 3lr + 3lr2 )
Note that this is done without regard to the width profile, and
setting the width to wnom . Now the energy of the actuator with
the extension is given as follows by substituting the terms in
(36) and the free displacement from (34) into (1):
3P(E3 )2 lwnom
Um,ext = L(lr ) (37)
8C44
The term L in the above equation is a unitless function of the
extension ratio:
(1 + 2lr )2
Fig. 8. Energy improvement as a function of the extension ratio. L(lr ) = (38)
(1 + 3lr + 3lr2 )
  Finally, assume that the extension adds negligible mass to the
C44 Fext l3 lext l2 actuator, and thus the energy and energy density are propor-
δf (l) = − + (32)
wnom 3 2 tional to L. The function L is plotted in Fig. 8 as a function
of the extension ratio lr .
Next note that the displacement at the distal end of the ex-
Note that if lr = 0, Eq. (37) reduces to the same en-
tension is a function of the displacement at the interface, the
ergy as for the rectangular case with no extension. To de-
slope at the interface, and the extension length.
termine the maximum improvement in energy, observe that
 limlr →∞ L = 4/3. Thus the maximum improvement in en-
dδ(x) 
δ(l + lext ) = δ(l) + lext (33) ergy and energy density is 4/3. Note that for all the equations
dx x=l
in Section 2.2, the extension is assumed to be perfectly rigid,
Adjusting the displacement terms in (31) and (32) for the or more practically to have a stiffness >10× that of the actu-
extension and simplifying by introducing the length ratio, ator without the extension (Figs. 9 and 10).
lr , defined as the ratio of the extension length to the actua-
tor length (without the extension), yields two displacement 2.3. Complete model
terms:
Combining the results from Sections 2.1 and 2.2 yields
Pl2 the complete description of the actuator performance. This
δp (l + lext ) = (1 + 2lr ) (34) is done by again performing the integrations on the curva-
2
ture, but now combining the width and extension terms in the
C44 Fext l3 external moment.
δf (l + lext ) = − (1 + 3lr + 3lr2 ) (35)
3wnom Fext (l(1 + lr ) − x)
Mx (x) = − (39)
To solve for the blocked force at the extension, the su- wnom ((2(1 − wr )/ l)x + wr )
perposition of the two displacements must be zero, i.e. The remainder of the procedure is identical to that of
δp (l + lext ) + δp (l + lext ) = 0. Using this and solving for Fext Section 2.2. First, the displacement due to the applied electric

Fig. 9. Energy improvement as a function of the geometry: (a) 3D plot of GU and (b) contour plot of the same function.
 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488 483

For convenience, call the term (1 + 2lr ) in (35) Gδ , the free

displacement geometry constant. Lastly, call GU the prod-
uct Gδ GFb ; thus the mechanical energy can be written as a
function of GU :
3P(E3 )2 lwnom
Um = GU (wr , lr ) (46)
8C44
Thus GU multiplies with the energy and also the energy den-
sity, therefore GU is a relative measure of the geometrically
improved energy density. Finally, the energy density is given
as:
3P(E3 )2 /8C44
DU =  GU (wr , lr ) (47)
n ρn tn
where ρn and tn are the densities and thicknesses of the nth
Fig. 10. Predicted effect of passive layer thickness on energy density for a layer respectively. The GU parameter is plotted below as a
bimorph. function of the width and extension ratios. This completes the
model for cantilever piezoelectric bending actuators includ-
field is found, along with the displacement due to an exter-
ing any number, anisotropy, or orientation of the constituent
nal force. These are then added and set to zero to find the
layers, number, placement, or makeup of piezoelectric layers,
blocked force. Note that the free displacement is identical to
and overall geometry.
the results in Section 2.2. First, integrating twice to find the
displacement from an external force yields the following:
C44 Fext l3 3. Fabrication
δf (l) = − Gl (wr , lr ) (40)
3wnom
From the design analysis, ultra-high modulus (UHM) uni-
The term Gl is a function of the width ratio and extension ratio directional carbon fiber composites are chosen for the pas-
Gl (wr , lr ) = (ga + gb )/gc where the gi terms are defined as sive layers, similar to [24]. Carbon fiber-based composites
follows: have the added benefit of being decent (≈ 1 × 10−3 5 cm)
ga = 6(wr − 1)(−3 − 2lr + 2wr + 2lr wr ), electrical conductors (for low current applications), thus no
  additional electrodes need to be implanted within the ac-
2 − wr
gb = 3(wr − 2)(−2 − 2lr + wr + 2lr wr ) ln , tuator layup. Ideally, such UHM materials would be used
wr to create the rigid extension. However, because of concerns
gc = 8(1 − wr )3 (41) for shorting the piezo electrodes, non-conductive fiber glass
layers are used. Since extension compliance is a significant
Recall again Eq. (33) for a rigid extension and substitute this concern the highest modulus glass material, unidirectional
into (40) yielding the following: S2Glass, is layered to stiffen the extension. Each layer is
laser-micromachined into desired shapes using a computer
C44 Fext l3
δf (l + lext ) = − Glext (wr , lr ) (42) controlled precision pulsed laser (QuickLaze from New Wave
3wnom Research Inc.) to control dimensions and relative ply angles.
where the parameter Glext (wr , lr ) = (gd + ge )/gc and the gi The composite material matrices are thermoset polymers ini-
terms are defined as follows: tially in a catalyzed uncured state (called prepreg) and are
cut in this state. After all materials are cut, the layers are as-
gd = 6(wr − 1)(3 + 4lr − 2wr − 4lr wr ), sembled with the desired layup and cured in a vacuum oven.
 
2 − wr The bonding between each layer is achieved by the flow of
ge = 3(−2 − 2lr + wr + 2lr wr ) ln
2
(43)
wr the composite matrix epoxy while curing. Thus, the actuator
requires neither electrodes nor additional bonding layers.
Now adding Eqs. (35) and (42) and setting this to zero yields To determine the geometric parameters for the desired ap-
the blocked force: plication, the effects of the various constituent parameters in
3P(E3 )wnom Section 2.3 on the output performance is determined. First,
Fb,ext = GFb (wr , lr ) (44)
2C44 l the width ratio and the length ratio are maximized to the
where the term GFb is called the blocked force geometry highest practical degree. For practical applications, the width
constant and is defined as follows: ratio will not reach the maximum value of 2 since a distal
end with zero width results in too fragile a structure, and the
(1 + 2lr )
GFb (wr , lr ) = (45) connection to the extension needs to be as rigid as possible.
Glext (wr , lr ) The length ratio cannot be increased indefinitely since the
484 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488

extension mass becomes a concern. The current design uses 3.2. Utilizing thermal expansion mismatches
a width ratio of 1.5 and an extension ratio of 1.
In choosing the geometry of the actuator, three output pa- Before curing, the materials in the actuator are free, and
rameters are used: the displacement, the blocked force, and when brought up to high temperatures for curing, they freely
the energy density. From the results in Section 2.3, the fol- expand or contract depending upon the sign of their CTE. The
lowing is clear: state before any significant cross linking of the matrix epoxy
occurs is referred to as the stress free state. Once bonding oc-
DU = g1 (tr ), δ = g2 (tr , l), curs, the materials are joined at the interface. By designing the
Fb = g3 (tr , l, wnom ) (48) elastic layer to have a proper CTE with respect to the piezo-
electric layer, a tailored stress is applied to the piezoelectric
where l and wnom are again the actuator length and nominal material after the actuator is subsequently brought down to
width respectively and tr the ratio of the passive layer thick- room temperature. Quantitatively, the thermal stresses de-
ness to a single piezoelectric plate thickness. First, for given veloped in each layer of the laminate can be determined by
constituent layer mechanical and piezoelectric properties, the examining Eq. (18). By solving (18) for the midplane strains
energy density is maximized over the passive layer thickness. and curvatures and noting that the strain in the ith layer is re-
From this plot it can be seen that the maximum energy density lated to these two terms via Eq. (50), the stresses and strains
occurs at a thickness ratio of approximately 0.35 for the bi- can be predicted on a ply by ply basis.
morph with respect to the given material properties and using
i = 0 + κzi (50)
unidirectional UHM composites. Note that the UHM energy
density is plotted alongside similar curves for other common Section 4.1 provides a diagram of the strains in the various
engineering materials for comparison reasons. Now the re- layers of the bimorph actuator.
maining two parameters, the displacement and blocked force,
have two unknowns, the length and the nominal width. Solv-
ing these two equations for application specific displacement 4. Driving techniques
and blocked force yields the length and width parameters.
The properties of the various engineering materials consid- There are a number of traditional methods for applying
ered for the passive and active layers are given in Table 2. a field to bimorph piezoelectric bending actuators. The first
two methods require only a single source which is connected
3.1. Piezoelectric material choice electrically in parallel or series with the electrodes of the bi-
morph as described by Wang et al. [17] and are shown in
One of the best criterion for choice of piezoelectric mate- Fig. 11(a). For the series case, the field is applied across the
rial is the strain energy density. The strain energy density is two electrodes and the polarization of each piezoelectric layer
defined as follows: is opposite. Thus the application of the electric field creates
  opposing strains in the two layers. For the case of the par-
1 1 2
emax = E (49) allel configuration, the center electrode is grounded, while a
ρ 2 max positive voltage is applied to the outer two electrodes of sim-
where emax is the maximum strain energy density, ρ the den- ilarly poled piezoelectric layers. Thus, instead of opposite
sity, E the Young’s modulus, and max the maximum strain. polarizations, the field orientation creates opposite strains.
This last term, max , can be thought of as derived from one of The major differences between these two methods are that
two different elastic modes: either the strain developed from the series configuration will require twice the voltage to ob-
external loading, or the induced piezoelectric strain at a given
field, namely dij Ei . Two piezoelectric materials are consid-
ered: PZT-5H, a soft polycrystalline ceramic, and PZN-PT,
a ferroelectric relaxor-based single crystal [10,23]. The for-
mer has the benefits of low cost, ease of availability, and
relatively high elastic modulus while the latter has a much
larger piezoelectric coupling coefficient with the drawbacks
of cost, availability, and poor fracture properties. The prop-
erties of both materials are given in Table 2. To determine
which strain to use in (49), choose min(d31 E3 , f ), where
f is the mechanical fracture strain. Thus, because of the ex-
tremely low fracture toughness associated with single crystal
materials, PZT shows a factor of two better performance.
Therefore, polycrystalline piezoelectric materials are used
throughout the following discussion. Fig. 11. Drive diagram for (a) single source and (b) dual source.
 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488 485

tain the same field, while the parallel method requires one
additional connection. The overall problem with these two
methods lies in the limit of the field that could be applied be-
fore depoling occurs within the layer that is poled antiparallel
to the field direction. For PZT at room temperature, this is
approximately 0.5 V␮m−1 , however the desired field (which
will draw the constituent actuator materials close to fracture)
is approximately 2–3V␮m−1 which would depole one of the
layers, making the actuator useless. This effect is amplified as
the operating temperature approaches the Curie temperature
or in the presence of applied mechanical stresses. It is clear
that in order to obtain the highest possible performance, each
piezoelectric layer must be kept under positive field with re-
spect to its polarization direction. Thus the achievable energy
Fig. 12. Comparison of hysteresis plots for all three dual source driving
density would be much too low for either of these techniques
methods (drive voltage is with respect to Vmax /2).
(as can be seen in Table 1).
The dual source drive schematics are shown in Fig. 11(b).
The first of these two is termed the alternating drive where the load cell, and measuring the resulting force. This gives a
the dual source drive either layer independently. Each source very linear curve for the blocked force as a function of field
is kept 180◦ out of phase with the other, each driven unipolar. as is shown in Fig. 13(a). To quantify the hysteresis in the
This keeps each layer driven only in a positive sense, however applied force, the load cell was put in contact with the free
requires two independent sources for each actuator, and thus end of the actuator and the bimorph was driven from Vd = 0
a total of 2n + 1 wires and 2n sources for n actuators. The to Vmax where Vmax = Vb for incrementally increased bias
second option, termed the simultaneous drive, instead biases fields. The results are shown in Fig. 13(b). It is well un-
the entire actuator. Thus the bias and ground can be com- derstood that piezoelectric materials undergo a softening at
mon for multiple actuators, requiring n + 2 wires and n + 1 higher fields. This softening will be apparent when observing
sources for n actuators. the Q and resonant frequencies as a function of the applied
field. As an example of this effect, the frequency response of
4.1. Results an unloaded bimorph was determined for iteratively increased
field magnitudes. Since the actuator alone is a high Q system,
The mechanical energy provided by these two competing this could not be tested up to the field magnitudes that will be
drive methods was measured by individually measuring the experienced when connected to a load. From the frequency
displacement and blocked force using custom built optical response seen in Fig. 14(a), the resonant frequency and Q can
[2] and strain [19] sensors respectively. First, each driving be extracted as a function of the field magnitude. Note that
method is evaluated based upon the static performance with the actuator is driven in simultaneous mode with Vmax = Vb .
no applied load. The results show that the alternating method The results are shown in Fig. 14(b). Finally, the performance
exhibits greater saturation, while providing approximately of the bimorph actuators described here are given in Table 3.
5% less displacement than the simultaneous drive for a given Note that in Table 3 there are two known causes of the dif-
maximum field. Under cyclic actuation however, the simul- ference between the predicted and experimental performance
taneous drive shows slightly greater hysteresis. These three values. The first is due to stress based d31 enhancements. It
differences can be seen by a comparison of the output traces is known that the piezoelectric coupling coefficients of poly-
in Fig. 12. Since for the simultaneous case the relative magni- crystalline piezoceramics is increased with applied tensile
tudes of the bias field and maximum drive field are variable, stress and decreased with compressive stress [1,8,22]. For
an attempt was made to compensate for this hysteresis in- the case of the bimorphs described here, the piezoelectric
crease by increasing the relative magnitude of the bias field. plate performing work is always in tension (either imposed
The idea is that the added hysteresis originates from stress de- elastically from the opposite plate or from an external re-
polarization of the piezoelectric plate which is experiencing action force) and will increase both peak displacement and
relatively less field magnitude during any given half stroke. blocked force. The second cause for the discrepancy is piezo-
The field at this point is less than is required to maintain po- ceramic softening under high drive fields (as is displayed in
larization while under high stresses. The results of this test, as the decrease in resonant frequency shown in Fig. 14(a)). This
shown in Fig. 12, display that this principle of reducing the softening will result in an increase in displacement and a de-
hysteresis does work, however the displacement amplitude crease in blocked force. Corrections for these effects will
decreases significantly. Next, the linearity and hysteresis in not be presented here, other than noting that altering d31
blocked force was examined using a custom built load cell and the elastic modulus values as a function of applied field
[19]. This was done first by incrementally applying a field to causes the predicted and measured performance to be nearly
a bimorph, optically zeroing the actuator displacement with identical.
486 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488

Fig. 13. Blocked force with the drive field (a) and force transducer measurements from bimorph showing hysteresis (b) both using the simultaneous drive
method (drive voltage is with respect to Vmax ).

Table 3 ing laminate plate theory for the design of bending actua-
Performance results for energy density optimized bimorph piezoelectric tors and sensors. The thermal, piezoelectric, external, and
bending actuators (for N = 32 actuators, simultaneous drive, 2.4 V␮m−1 )
total strains (because of the principle of superposition) are
Parameter Units Predicted Measured Error (%) shown for the outer surface of the top piezoelectric layer
δa µm 406 520 ± 56.9 +28.1 (because of symmetry) in Fig. 15 for a bimorph as well
Fb a mN 136 123 ± 20.5 −9.6 as for the passive layer for an externally applied 100 mN
m mg 11.72 11.75 ± 0.8 +2.6
Um ␮J 28 32 ± 7.1 +14.3
load.
DU J kg−1 2.35 2.73 ± 0.5 +16.2 Note that in Fig. 15 the total strain is nearly constant along
a Peak-to-peak. the length of the actuator and close to the fracture strain of
the material, thus ensuring that each element of the piezo-
5. Discussion electric material is performing nearly maximal useful work.
In summary, there are four techniques described here to im-
At the beginning of this paper, the concept of high energy prove the energy density of piezoelectric cantilever bending
density actuators was introduced as structures in which all actuators: width tapering, extension, high performance ma-
active materials are driven as close as possible to their maxi- terial choice, and high field drive. While the geometric al-
mum achievable strain (limited by either saturation, fracture, terations contribute an increase of 33% (Fig. 9), the use of
breakdown, etc.). It is desirable at this point to estimate nu- composite materials yields between a maximum of 5–10%
merically the strain present in each layer of the actuators due improvement (depending upon material choice, Fig. 10) as
to internal (thermal expansion, piezoelectric displacement) well as other ancillary benefits. Finally, the dual source si-
and external (external loads) excitations typically expected multaneous drive technique coupled with the strain unifor-
for the given application. This is readily accomplished by mity provided by the geometric modifications provide an
calculating the midplane strains and curvatures when sub- increase of approximately a factor of 10. In total, such ac-
stituting the internal and external forces and moments into tuators described here exhibit energy densities on the order
Eq. (18) and then applying this to each layer via Eq. (50). of 10–50 times those of commercially available adaptations
This ease of strain analysis is an ancillary benefit of us- (Fig. 16).

Fig. 14. Bimorph frequency response for varying field magnitudes (a) and the associated resonant frequency and Q (b).
 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488 487

Fig. 15. Strains in the constituent layers of the bimorph actuator (simultaneous drive).

[8] S.W. Meeks, R.W. Timme, Effects of one-dimensional stress on piezo-

electric ceramics, J. Appl. Phys. 46 (10) (1975) 4334–4338.
[9] J.-D. Nicoud, J.-C. Zufferey, Towards indoor flying robots, in: Pro-
ceedings of the IEEE/RSJ International Conference on Intelligent
Robots and Systems, Lausanne, Switzerland, October 2002, pp. 787–
Fig. 16. Completed bimorph actuator. 792.
[10] S.E. Park, T.R. Shrout, Ultrahigh strains and piezoelectric behavior in
relaxor based ferroelectric single crystals, J. Appl. Phys. 82 (4) (1997)
Acknowledgments 1804–1811.
[11] J. Pasquero, V. Hayward, STReSS: a practical tactile display system
The authors would like to thank Jacoby Hickerson for as- with one millimeter spatial resolution and 700 Hz refresh rate, in: Pro-
sistance in actuator construction and Srinath Avadhanula for ceedings of the Eurohaptics, 2003, pp. 94–110.
[12] M. Sitti, D. Campolo, J. Yan, R.S. Fearing, T. Su, D. Taylor, T.D.
insightful discussions. This work was supported under fund- Sands, Development of PZT and PZN-PT based unimorph actuators
ing by ONR MURI N00014-98-1-0671, DARPA and NSF for micromechanical flapping mechanisms, in: Proceedings of the IEEE
IIS-0083472. International Conference on Robotics and Automation, Seoul, Korea,
May 2001.
[13] J.G. Smits, A. Ballato, Dynamic admittance of piezoelectric can-
tilever bimorphs, J. Microelect. Mech. Syst. 3 (3) (1994) 105–
References 112.
[14] J.G. Smits, W.S. Choi, The constituent equations of piezoelectric het-
[1] M. Algueró, B.L. Cheng, F. Guiu, M.J. Reece, M. Poole, N. Alford, erogeneous bimorphs, IEEE Trans. Ultrasonics Ferroelect. Freq. Contr.
Degradation of the d33 piezoelectric coefficient for PZT under static 38 (3) (1991) 256–270.
and cyclic compressive loading, J. Eur. Ceram. Soc. 21 (2001) 1437– [15] E.B. Tadfor, G. Kósa, Electromechanical coupling correction for piezo-
1440. electric layered beams, J. Microelect. Mech. Syst. 12 (6) (2003).
[2] S. Avadhanula, R.J. Wood, E. Steltz, J. Yan, R.S. Fearing, Lift force [16] Q.M. Wang, L.E. Cross, Constitutive equations of symmetrical triple
improvements for the micromechanical flying insect, in: Proceedings layer piezoelectric benders, IEEE Trans. Ultrasonics Ferroelect. Freq.
of the IEEE/RSJ International Conference on Intelligent Robots and Contr. 46 (6) (1999) 1343–1351.
Systems, Las Vegas, NV, October 2003. [17] Q.M. Wang, X.H. Du, B. Xu, L.E. Cross, Electromechanical cou-
[3] D. Campolo, R. Sahai, R.S. Fearing, Development of piezoelectric pling and output efficiency of piezoelectric bending actuators, IEEE
bending actuators with embedded piezoelectric sensors for microme- Trans. Ultrasonics Ferroelect. Freq. Contr. 46 (3) (1999) 638–
chanical flapping mechanisms, in: Proceedings of the IEEE Inter- 646.
national Conference on Robotics and Automation, Taipei, Taiwan, [18] M.S. Weinberg, Working equations for piezoelectric actuators and sen-
September 2003. sors, J. Microelect. Mech. Syst. 8 (4) (1999) 529–533.
[4] D.L. DeVoe, A.P. Pisano, Modeling and optimal design of piezoelectric [19] R.J. Wood, R.S. Fearing, Flight force measurements for a microme-
cantilever microactuators, J. Microelect. Mech. Syst. 6 (3) (1997) 266– chanical flying insect, in: Proceedings of the IEEE/RSJ International
270. Conference on Intelligent Robots and Systems, Maui, HI, October
[5] M. Gogola, G. Fischer, M. Goldfarb, E. Garcia, The development of 2001.
two piezoelectrically-actuated mesoscale robot quadrupeds, in: Pro- [20] R.J. Wood, S. Avadhanula, M. Menon, R.S. Fearing, Microrobotics
ceedings of the SPIE Conference on Microrobotics and Microassembly, using composite materials: the micromechanical flying insect thorax,
vol. 3834, Boston, MA, September 1999, pp. 76–84. in: Proceedings of the IEEE International Conference on Robotics and
[6] W.E. Green, P.Y. Oh, An aerial prototype for situational awareness in Automation, Taipei, Taiwan, September 2003.
closed quarters, in: Proceedings of the IEEE/RSJ International Confer- [21] J. Yan, R.J. Wood, S. Avadhanula, M. Sitti, R.S. Fearing, Towards flap-
ence on Intelligent Robots and Systems, Las Vegas, NV, October 2003, ping wing control for a micromechanical flying insect, in: Proceedings
pp. 61–66. of the IEEE International Conference on Robotics and Automation,
[7] H. Kioua, S. Mirza, Piezoelectric induced bending and twisting of lam- Seoul, Korea, May 2001.
inated composite shallow shells, J. Smart Mater. Struct. 9 (2000) 476– [22] G. Yang, S.-F. Liu, W. Ren, B.K. Mukherjee, Uniaxial stress de-
484. pendence of the piezoelectric properties of lead zirconate titanate
488 R.J. Wood et al. / Sensors and Actuators A 119 (2005) 476–488

ceramics, in: Proceedings of the 12th IEEE International Sym- his MS in EE from U.C. Berkeley (2001) and BS in EE from Syracuse
posium 2000 on Applications of Ferroelectrics, 2001, pp. 431– University (1988).
434.
[23] J. Yin, B. Jiang, W. Cao, Elastic, piezoelectric, and dielectric properties Erik Steltz is a third year graduate student studying towards his PhD in
of 0.995Pb(Zn1/3 Nb2/3 )O3 –0.45PbTiO3 single crystal with designed the Department of Electrical Engineering and Computer Sciences at the
multidomains, IEEE Trans. Ultrasonics Ferroelect. Freq. Contr. 47 (1) University of California, Berkeley. He received a BS in Engineering Science
(2000) 285–291. from the Pennsylvania State University (2002). He has published in the area
[24] K.J. Yoon, S. Shin, H.C. Park, N.S. Goo, Design and manufacture of a of electromagnetic behavior of thin films, but his current research focuses
lightweight piezo-composite actuator, J. Smart Mater. Struct. 11 (2002) on biologically inspired robotics and millirobotics.
163–168.
Ronald Fearing is a professor and vice chair for undergraduate matters in
the Department of Electrical Engineering and Computer Sciences at Univer-
Biographies sity of California, Berkeley, which he joined in January 1988. His principal
research interests are in micro robotics, including mobile micro-robots, dex-
trous micromanipulation and teletaction. He has a PhD from Stanford in EE
Robert Wood is a PhD candidate in the Department of Electrical Engineer-
(1988) and SB and SM in EECS from MIT (1983). He received the Presi-
ing and Computer Sciences at the University of California, Berkeley. His
dential Young Investigator Award in 1991, and is the co-inventor on 2 US
research interests include biomimetic systems, microrobotics, micro air ve-
patents.
hicles, and composite materials for microrobotic applications. He received