Recent Progress in Materials Issues for Piezoelectric MEMS


  • Paul Muralt

    Corresponding author
    1. Laboratoire de Céramique, Ecole Polytechnique Fédérale de Lausanne EPFL, Lausanne, Switzerland
      †Author to whom correspondence should be addressed. e-mail:
    Search for more papers by this author

  • D. Green—contributing editor

†Author to whom correspondence should be addressed. e-mail:


Piezoelectric materials play a crucial role in a large number of devices and applications modern society would not like to miss. Mobile phones and ultrasonic imaging are just the most prominent ones. Since two decades, miniaturization of mechanical devices in silicon technology is a major research direction in engineering known under name of MEMS, which stands for micro-electro-mechanical systems. Piezoelectricity fits very well into this concept and was expected right from the beginning to play its role in MEMS. The breakthrough was made with RF filters in mobile phones working on the principle of standing thickness waves in AlN films. What counts here is acoustic quality and stability. The force champion among piezoelectric thin film materials, Pb(Zr,Ti)O3 gave more problems in processing, and requires more patience to meet requirements and needs for a mass applications. It seems, however, that the breakthrough is imminent. This article attempts to give an overview of the field, highlighting recent achievements, introduce operation principles, and describe some applications.

1. Introduction

Micro-electro-mechanical systems (MEMS) were born as a new technological discipline during the 1980s (for an introductory text book see Malouf1). The idea of the pioneers was to enlarge capabilities of integrated circuits based on silicon beyond pure electronics by adding mechanical elements, which were made of silicon and further materials of semiconductor technology. The addition of mechanics extended the application range of silicon technology to motion sensors, pressure and force sensors, small actuators, and a number of acoustic and ultrasonic devices, most importantly resonators for signal treatment. In order to profit from symbiosis with electronics, those mechanical elements should, of course, be controlled by electronic signals. Evidently, this new silicon technology makes only sense for small, miniaturized devices. The technical advantage comes from the fact that powerful thin film deposition and patterning techniques as used for semiconductor fabrication allow for unprecedented precision of mechanics in the nano to micrometer range. As a large number of devices are produced in parallel on the same wafer (batch processing), the cost level is acceptable in spite of expensive fabrication tools, at least at high-production volumes. Concerning processing, the chemistry of silicon turned out to be very helpful: high-etching rates of anisotropic wet etching in a base solution (e.g., KOH), and anisotropic deep silicon etching in a plasma reactor are crucial issues in efficiently tailoring silicon. Over the last 20 years, MEMS technology became a proven and mature technology with many applications. While “MEMS” is still taken as a standing brand name for the field, the actual MEMS field has become much wider than stipulated by the notion of electro-mechanics, including thermal, optical, magnetic, chemical, biochemical, and further functional properties. Also the main material of the device is not necessarily silicon, but may be glass or plastics, especially for bio-medical applications.

It is expected that piezoelectric materials with their intrinsic electro-mechanical coupling should and will play an important role in MEMS. Indeed, piezoelectric thin film devices based on ZnO were among the first ones to be demonstrated.2 In piezoelectric materials, a deformation occurs upon application of an electric field (converse effect) whereby the obtained strains depend to first approximation linearly on the electric field. The opposite effect works as well: Upon application of a stress, charges develop on the surfaces of the piezoelectric body (direct effect). The most frequently applied bulk material is PZT, a solid solution with formula PbZrxTi1−xO3, which exhibits outstanding piezoelectric properties at the morphotropic phase boundary at x=0.53. Considering the applications of bulk materials (including thick films), we identify many typical features reflecting their unique characteristics, which may be also unique on the micro or nanoscale, and also useful for applications at small dimensions. The use of piezoelectrics in scanning microprobe microscopy is justified by the high rigidity prevailing during deformation, and enabling sub-nanometer precision in all directions. Piezoelectrics are used to generate and sense ultrasonic waves as needed in ultrasonic imaging at frequencies of 1–50 MHz, and in nondestructive testing at similar frequencies, by virtue of the large efficiency in energy conversion between electrical and mechanical energy. In signal filtering, piezoelectric resonators play a dominant role in the range 0.1–10 GHz thanks to their high-quality factors, outperforming by far LC resonators. In injection valves for combustion motors, piezoelectrics are applied thanks to large forces and high operation speeds. Without discussing dimensional details one can summarize advantages as follows:

  • 1strong forces, or alternatively large excursions in bending structures;
  • 2low voltage because of high dielectric constants;
  • 3high efficiency in energy conversion, and equivalently low noise detection;
  • 4high-speed and high-frequency operation possible;
  • 5high acoustic quality of many piezoelectrics (not true in multi-domain ferroelectrics);
  • 6and last but not least, a rather linear behavior.

For thin-film applications, the same principles remain valid; however, one has to keep in mind that smaller dimensions lead on the one hand to smaller voltages, on the other hand to weaker forces, smaller excursions, and higher resonance frequencies. Piezoelectricity is not the only possible choice. There are also other electro-mechanical phenomena available in the microworld, and not necessarily the same as in the macroworld. There is first of all electrostatic interaction across an air or vacuum gap of a capacitor3,4 or between comb fingers.5,6 Even with the disadvantage of nonlinearity, electrostatic interaction can be well used for actuators, and with a bias voltage, also for sensors and transducers. Not so practical in the macroworld (very high voltages needed), electrostatics can be a smart solution in the microworld. In theory, electrostatic devices may even reach an electro-mechanical coupling coefficient (or conversion efficiency) k2 of 100% in the small signal limit,4 a value that hardly exceeds 60% in piezoelectrics. A further competing effect is the thermo-mechanical bimorph effect. It can be used for actuation if a microheater is embedded in the device. The heat dissipation, and thus power consumption of such a device is naturally higher, the application frequency lower, than with a piezoelectric drive. The same holds for shape memory alloy actuators. Further possibilities are offered by electromagnetic interactions, and magnetostrictive materials. Disadvantages are in general resistive losses in coils. On the other hand, magnetic interaction can be an advantage if remote power feeding is required.7

The scientific community was always aware of the importance of piezoelectric MEMS. Interestingly, PZT thin films and their integration were developed in parallel to general MEMS processing. Both fields profited from a general improvement of physical and chemical deposition techniques during the 1980s, and dry etching techniques during the 1990s. However, the main momentum in ferroelectric films came from memory applications (ferroelectric random access memory, FERAM8,9). While PZT FERAMs have now found their niche markets and appear to be more and more established in low-power/low-voltage nonvolatile memories, PZT MEMS have not yet an established place in MEMS. The implied industry is much smaller, and even if physical criteria speak for PZT MEMS, the availability of high-quality material, and its price play an important role. Whereas AlN has become well available due to numerous companies and institutes working on thin film bulk acoustic wave RF filters—the first piezo-MEMS success—the ferroelectric Pb(Zr,Ti)O3 thin film is hardly available on a commercial basis. This is not only due to the very limited number of experienced producers (producing mostly for themselves only), but also due to the integration scheme that is often not well mastered and which imposes its rules to the design.10 As a result, PZT MEMS is only considered if there is no other, less demanding solution. It seems now that there is at least one good mass application justifying the effort in PZT MEMS: ink jet printers.

In this feature article, it is attempted to give an overview on current status and further research issues in ferroelectric thin films for piezoelectric MEMS applications. Focus is given on the recent progress, avoiding repetition of earlier reviews.11,12 A good indicator of the progress is the published value of the effective in-plane transverse piezoelectric coefficient e31,f, as measured through D3=e31,f (S1+S2) of a bending structure in parallel plate geometry (Fig. 1). D3 is the displacement field (charge Q3 per electrode area), and S1 and S2 are the in-plane strains. This coefficient is very relevant for MEMS applications, indicates at the same time a good density, and is conveniently measured in the direct13–16 and converse mode.17 In Fig. 2, the reported e31,f is shown as a function of the year of publication. A constant progress is observed in the case of PZT thin films integrated on silicon. The major integration issues and texture control issues were solved in the 1990s. It is instructive to note that the performance of integrated films on silicon now outperform the epitaxial films, which were first better, because in their case no integration and seeding issues had to be solved. The lead-based epitaxial films always show the same values in the range of −11 to −13 C/m2, whether it is a PZT, or a relaxor PMN–PT (Pb(Mg,Nb)O3–PbTiO3).

Figure 1.

 Simple beam structure with parallel plate capacitor geometry.

Figure 2.

 The history of the transverse piezoelectric coefficient e31,f as an indicator of the advancement in the field. The films are in the 0.5–2 μm thickness range. PZT thin films integrated on silicon (data from Shepard and colleagues)14,15,17,30,39,108,109 are compared with several types of lead-based epitaxial films (PZT) 110,111, BiScO3–PbTiO3,112 Pb(MgNb)O3–PbTiO3,113–115 relaxor type integrated thin films such as Pb(Yb,Nb)O3–PbTiO3.116–118 The values derived from ceramics data refer to early PZT ceramics (Jaffe), standard hard PZT 4 (product of Morgan Electroceramics), and modern optimized PZT ceramics, the 3203 of Motorola. The small numbers aside of the points are the film thickness in micrometers.

The difference between integrated and epitaxial films cannot be explained by a thickness effect, as both series contain values in the 2–3 μm range (see Fig. 2). Thickness would be a relevant parameter, because it was indeed observed that thicker films show more domain contributions (see Section V).18,19 Nevertheless, the explanation has to do most likely with these contributions (if we exclude simple process quality reasons). A value of −10 C/m2 is about the expected value for pure intrinsic piezoelectricity, meaning the response of a single c-domain (according to accepted values for electrostrictive, elastic, and dielectric constants of the lattice for PZT40/60 20,21). So one could argue that the lattice in epitaxial films is perfect that, however, the ferroelastic domains do not cooperate as well as in the best films grown on platinized silicon. There is a difference in domain configuration due to different thermal strain histories, because the substrates for epitaxy (MgO, SrTiO3, etc.) exhibit larger thermal expansion coefficients than silicon. As a result, the a-domain volume fraction in tetragonal phases is much larger in a film on silicon than in the others (see discussion in Section V). A further difference lies in the electrode: Pt as used for integrated films is a true metal that can screen (or compensate) the polarization charges on a very short range by free electrons. This is quite different when using an oxide electrode. Charge compensation involves a deeper zone, and might even involve oxygen vacancies. Furthermore, the nature of chemical bonds across the interface is different. Stress driven Pt diffusion at the interface is known to happen at the growth temperature.12 All such effects could lead to differences in dislocation densities and strain relaxation close to the interface, and thus eventually to stronger ferroelastic domain wall pinning near oxide electrodes, which would impede domain wall contributions to piezoelectricity at small signal response level.

Relaxor materials do not seem to give an advantage over PZT. This can be explained by: (i) Relaxors are more difficult to nucleate and to grow, thus requiring higher growth temperatures. This complicates integration tasks. (ii) The still relatively low growth temperatures (as compared with bulk materials) do not allow for a very homogeneous distributions of the disordered B-site ions, leading to a much lower dielectric constant (see Section II). (iii) The piezoelectric eij coefficients express mechanical stress per applied electric field, and are thus proportional to the stiffness. The high-strain response of relaxors does not equally translate in stress response because relaxors are softer.

Theoretically, well-oriented or textured films should exhibit superior properties than ceramics. This is indeed achieved comparing with pure PZT ceramics (such as the old material published by Jaffe and Berlincourt22), and with standard PZT 4 (Morgan Electroceramics). However, there are modern ceramics with elaborate substitutions and dopants that still are superior to the best PZT thin films (Motorola's 3203 in Fig. 2). One can thus argue that there is room for improvements.

II. Processing of Lead-Based Ferroelectric Thin Films

Much of Pb(ZrxTi1−x)O3 (PZT) processing was learned during FERAM development.23,24 PZT and related materials can be deposited by the major standard methods applied in wafer processing, i.e., metal–organic chemical vapor deposition (MOCVD), sputtering, and chemical solution deposition (CSD). The latter includes the sol–gel techniques. Common to all methods are the pure material phenomena such as nucleation. The perovskite structure is sufficiently complex to exhibit a so-called nucleation limited growth—at least if we do not deal with homoepitaxy. Hence, substrate structure and chemistry as well as chemical composition of the first monolayers play an essential role (see Muralt25 for details). It is for instance observed that nucleation on Pt(111) needs some lead excess, while no excess is needed for continuing growth on an existing PZT film. Another example is the orientation as obtained on Pt(111), which depends on the chemical composition of the first few monolayers. Starting a layer of PbTiO3 with TiO2 leads rather to (111) orientation, starting with PbO leads rather to (100)-orientation.26 Such behavior allows for a texture control.27 On very small TiO2 islands, quasi single crystal platelets were obtained, allowing for a site controlled, or registered growth of ferroelectric nanosized structures.28,29 Lead excess is an important parameter during growth and crystallization anneal. The higher volatility of PbO requires an excess either during growth, or in as-deposited films before post-annealing as required in techiques such as sol–gel. Such techniques might lead in addition to gradients in Zr/Ti ratio. Such gradientes can be reduced if they are accounted for properly.30

Crystalline phase nucleation is a crucial issue in two respects: First of all it is important to achieve the right crystalline phase of the material. Secondly, for property optimization, the crystalline orientation or texture must be a specific one and needs to be reproducibly controlled. The growth substrate plays of course a very prominent role. It is essential to start with well-defined substrates, meaning bottom electrodes in many cases. A standard solution is a passivated Pt film. This means that the adhesion layer (based on Ti, Ta, Zr, etc.) is oxidized before the deposition of the ferroelectric oxide.31,32 Alternatively, oxide conductors can be used, such as SrRuO3, or LaNiO3.33 Of course one has to check if the resistance of the electrode is compatible with the application. Perovskite phases are already enough complicated to show nucleation-limited growth. This means that the activation energy for nucleation is higher than the one for growth. It can be demonstrated that the phase nucleated during the first nanometers of film growth, i.e., perovskite or pyrochlore will prevail until the end of film growth, even if the processes are identical after the first monolayers. One of the reasons is the volatility of the Pb or PbO species. It turned out that when using Pt bottom electrodes, affinity, seed layers or template layers are very helpful. Evidence for this is shown in Fig. 3. In this experiment, a 2-nm-thick TiO2 (rutile(100)) layer on top of an epitaxial Pt film (SrTiO3(111) substrate) was patterned into islands by e-beam lithography.28 Subsequently, a 20-nm-PZT film is grown by in situ sputtering at around 600°C. The upper half of the figure shows a large 22 μm island. The nuclei density on the latter is about 60 times larger than on bare Pt(111). Near the island, the nucleation density is even lower than far from the island. This means that PbO adatoms are gettered by the island and its growing nuclei. A diffusion distance of about 1 μm is evidenced. On very small islands of 150 nm × 150 nm, single grains of PZT(100) were obtained (lower half of figure), enabling a registered growth of PZT nanoislands. On large areas, the Ti seed layer leads to (111) orientation.27 However, on the very small islands, (100)-oriented grains may grow because of the large PbO flux arriving by lateral diffusion (this topic is discussed in more detail in Muralt).25

Figure 3.

 Condensation of PbTiO3 (nominally 2 nm) on which 18 nm of PZT40/60 is grown. The whole structure was grown on a 100 nm epitaxial Pt(111) film on a SrTiO3 single crystal. (a) The 2 μm × 2 μm large square in the center was originally coated by 2 nm TiO2. The highlighted area indicates the depletion zone around the affinity square.28 (b)Nucleation on small affinity spots. The inserts show the piezo-AFM response with amplitude (left) and phase (right) image (from).28

If we do not deal with a uniform substrate, for instance with TiO2 particles that form in grain boundaries due to diffusion from a Ti adhesion layer, an irregular nucleation of all orientations may result. One possible solution for PZT(111) is thus a 2-nm-thick TiO2 rutile seed layer, or alternatively provoke Pb–Pt alloying by suitable annealing conditions, and promote in this way (111)-orientation.34,35 For homogeneous nucleation of the (100)-orientation, it turned out to be very useful to use a PbTiO3 template layer that grows in (100)-orientation on Pt (100) if enough lead flux is assured.36 PbTiO3 nucleates much more easily than PZT 50/50, and in addition does not exhibit the competing pyrochlore in the temperature range of interest.

Diffusion phenomena do not only occur in in situ growth, they are also important in postannealing of amorphous phases, as required in sol–gel techniques. The existence of such a gradient was sometimes postulated to explain the moderate peaking of piezoelectric and dielectric properties at the morphotropic phase boundary,37 was once observed by Auger depth profiling,38 and also revealed by etching profiles in PZT MEMS structures.39 PZT is a solid solution system existing in the whole compositional range in the form of the perovskite structure. The free energy of formation is significantly more negative (more exothermal) with Ti-rich compositions. This means that in the growth front between amorphous and perovskite phase the Ti ions are more likely captured by the crystalline phase than the Zr ions. At a growth temperature of 650°C, the capture probability was estimated to be 56% for Ti and 44% for Zr.30 It turns out that the diffusivity in the amorphous phase is quite high as compared with the propagation speed of the crystallization front. As a consequence, a gradient of composition is established in the amorphous phase leading to a Ti flow to the growth front, and a Zr flow away from the growth front (see Fig. 4). Within the perovskite phase there is no diffusion of cations anymore. This is a distinct difference between bulk processing at around 1200°C, and thin film processing at 600° to 700°C. The gradient can be reduced by a factor 5 if in each individually annealed layer a counter gradient is installed in the amorphous phase, anticipating the problem.30 One can use for instance for spins of different solutions per one crystallization anneal. The first solution must contain more Zr, the last more Ti than the target composition. Figure 4 shows dark TEM images comparing gradient, and the gradient-free films. The contrast arises from the fact that at the morphotropic phase boundary composition (53/47) the symmetry changes from tetragonal (Ti-rich) to rhombohedral (Zr-rich). In case of the (standard) gradient film we deal with a well visible tetragonal/rhombohedral phase mixture. The compositional fluctuation was quantitatively measured by EDAX. The Zr concentration in the gradient film fluctuates between 41% and 65%, in the gradient-free film between 51.5% and 55.5%. The removal of the gradient made the dielectric constant rise from 1300 to 1700. In addition, the piezoelectric properties were increased considerably (Fig. 5). The peaking at the morphotropic phase boundary is now more pronounced.

Figure 4.

 Dark field TEM images of 2-μm-thick PZT53/47 films deposited by a sol–gel technique in eight steps. Each step includes four spun and pyrolysed layers followed by a crystallization anneal. The image contrast arises from the (201) diffraction from a film of homogeneous 53/47 solution (a), and from a film synthesized with four different solutions anticipating preferential Ti capture for perovskite formation in order to achieve a homogeneous material (b).30 The schematic to the left depicts the formation of the concentration gradient during crystallization anneal per one single annealed layer of 250 nm.

Figure 5.

 Improved peaking of the transverse e31,f coefficient at the morphotropic phase boundary by increased film thickness and improved homogeneity of film composition. Combined results from Calame and colleagues.30,39,119

Aside of this concentration issue, sol–gel films of PZT are very well densified during annealing. This is not only seen on TEM or SEM cross sectional images, it is also manifested in the very high e31,f coefficient of up to 17 C/m2. The material flow is quasi one-dimensional along the z-direction, and allows for an unperturbed columnar growth. The reason is that we deal with the above mentioned nucleation limited growth, which as a consequence leads to nucleation at interfaces, surfaces, and particles (hetero nucleation) before nucleation within the amorphous phase can take place (homo nucleation). For this reason is it so important to promote nucleation by a suitable substrate, and to work particle free. When PZT films are sputter deposited at lower temperatures, and postannealed later as in case of sol–gel films, the density seems to be never as good. It appears that during sputtering already the material becomes inhomogeneous by nanocrystallite and roughness formation. The densification is never as homogeneous, leading finally to pores. Sputtering leads to good results only if the film is deposited in situ, meaning at high temperature (about 600°C) directly in the perovskite phase.40

Homogeneous densification does not occur in all materials so easily as in sol–gel PZT. In relaxor-type materials there are usually much more disturbing second phases. The related difficulties are well traced in the Pb(Sc0.5Ta0.5)O3 (PST) system.41–43 The point there is that the first nucleating phase is a Pb–Ta–O pyrochlore, and that there is no stable Pb–Sc–O second phase (Fig. 6). As a consequence, the pyrochlore has to be “remelted” to form later the PST perovskite, leading to a higher annealing temperature. In addition, not all the lead finds place in the Ta pyrochlore and thus evaporates partially as PbO. The final structure becomes PbO deficient, meaning oxygen vacancies in Sc3+-rich zones, and Pb vacancies in Ta5+-rich zones. This type of disorder is different from the disorder obtained by quenching ceramics from 1200°C, and not surprisingly, the observed relaxor behavior does not show the large peak in dielectric constant as expected. However, one can refill the missing PbO at 900°C in a saturated PbO/O2 atmosphere, and obtain much improved properties.42 Unfortunately this is not a viable approach for MEMS fabrication. The above-described problem of materials inhomogeneity and its impact on relaxor properties was much earlier already described for the case of fine grained and porous ceramics.44

Figure 6.

 Proposed schematic energy for PST.42

III. Microstructure—Piezoelectric Coefficients—Operation Principles

Thin film microstructure and the electrode system must fit to the operation mode. In this section the various electrode geometries are discussed. The piezoelectric film interacts through the mechanical parameters strain (Si) and stress (Ti) with the other parts of the device and external forces. The piezoelectric effect relates the strain tensor Si (i=1–6, reduced notation45), or the stress tensor Ti (i=1–6) with the vectors of the electric field Ei or the electric displacement field Di (i=1–3). These relations are linear and are written as follows when using the electric field as electrical variable:


In 10 out of 20 piezoelectric crystal classes there exists a polar axis. All the piezoelectrics used in thin film form belong to this material group. It is conventional to assign the coordinate index 3 to the polar axis direction, and to define d33 as positive quantity.

In MEMS technology, most of the piezoelectric thin films are polycrystalline materials. The piezoelectric effect is averaged over all the grains. At this point one has to distinguish between ferroelectric (such as PZT) and nonferroelectric polar materials (such as AlN and ZnO). The latter do not allow reorientation of the polar axis. In this case, the material growth process has to provide for a textured structure that includes the alignment of the polar directions. In ferroelectric materials, the polar axis can be reoriented by an electric field. Although the ideal ferroelectric can be easily switched at room temperature (a property that is used in memories), the ferroelectric thin film is usually better poled (i.e., more remnant polarization) when rising the temperature 100°–200°C for moving charged defects, especially oxygen vacancies, in order to un-pin domain walls and comply defect positions with the domains of the poled material.46 To some degree, the charged defects provoking back switching can also be compensated by electrons and holes generated by UV illumination during poling.47

Many of the phenomena during poling are not yet fully understood. For the moment we just state that at the end of poling, the polarization is aligned as much as possible with the poling field. In ceramics, it is conventional to assign the index 3 to this poling direction. For the case of the vertically stacked capacitor structure (see Fig. 1), direction 3 is perpendicular to the film plane, using the same convention as for polycrystalline ferroelectrics. The piezoelectric coefficient d33 thus describes the piezoelectric effect in the out-of-plane direction, as e.g., the thickness change of the free film. The directions 1 and 2 are therefore in the plane of the film, and usually equivalent. The transverse in-plane strains due to E3 are isotropic and d31 and d32 are the same. The polycrystalline thin film geometry thus recovers in this case the cylindrical symmetry of poled ceramics, yielding the same nonvanishing elements of the piezoelectric tensor as for tetragonal symmetry, i.e., d33, d31, d15. In the case of nonferroelectric materials, the polar crystal axis is by convention the “3” axis as well. In the parallel plate capacitor, a maximal longitudinal response (d33) is obtained with the film orientation (0,0,1) or (0,0,−1). Uniform texture of either of the two possibilities has thus to be achieved. In Fig. 7(a), the arrows symbolize the direction of the polar axis in a nonferroelectric system such as AlN and ZnO, or the average polarization per grain in a poled ferroelectric material. The in-plane directions are again equivalent if the film is polycrystalline, or if the polar axis is a threefold or higher rotation axes. In both cases the resulting symmetry is thus again the same as for poled ceramics. Polar films also offer the possibility to work with shear strains d15 contribution. Interdigitated electrodes48 can be applied for this purpose (see Fig. 7(b)–(c)). Between the electrodes the field is more or less parallel to the film plane, i.e. perpendicular to the polar axis of a well-grown polar film. In this region, inline image yields an alternating compression and dilatation of the surface layer. This mode is mainly used to excite surface waves in RF filter applications. Shear coefficients are more difficult to access in ferroelectric films, as the polarization direction follows always the poling field, which has the same geometry as the operating field when the electrodes are the same. In this case Fig. 7(d) does not apply. Instead, between the electrodes there is always a positive strain as the E-field is parallel to the local polarization. This situation is described in Fig. 8. This mode can be used as alternative to the parallel plate capacitor structure for bending actuation, or bending strain measurement. In actuators, a larger deflection effect was measured in this configuration.49 In energy scavenging, the use of interdigitated electrodes in conjunction with ferroelectrics is practically necessary because the harvested voltage must be as large as possible (see later).

Figure 7.

 Electrode systems for driving piezoelectric films: (a) Planar capacitor structure with top and bottom electrode. (b) Electric field with interdigitated electrodes. (c) Case of polar film with uniform perpendicular polarization. (d) Schematic drawing of strains induced by d15 between the electrodes, and d33, d31 below the electrodes for situation in (c).

Figure 8.

 Graph showing electric field and polarization when ferroelectric thin films is used in conjunction with interdigitated electrodes.

Sometimes interdigitated top electrodes are combined with an extended bottom electrode. In case of stable polar films (AlN, ZnO) the latter can be floating at half of the potential between the top electrodes. The electric field is vertical to the film plane like in a parallel plate capacitor. However, the polarity is alternating, and thus the in-plane piezoelectric stress. This type is also useful to excite surface waves or Lamb waves, sometimes even with larger coupling factor than without floating bottom electrode. For ferroelectric films this mode is less effective, because after poling with the same electrodes, the electric field is parallel to the electric field below each finger electrode (the wavelength is half of what it is with polar films). Nevertheless some nice results on Lamb wave devices were obtained with this geometry.50

IV. Effective Piezoelectric Coefficients

The main difference between thin film and bulk materials lies in the fact that thin films are used in a composite structure, where the total elastic properties are often dominated by the other part of the structure. This other part may be a silicon cantilever, a silicon oxide or a nitride membrane, for instance. The interaction with the substrate is very anisotropic. At the interface along the in-plane directions (indices 1,2), the piezoelectric thin film and the substrate have identical strains (S1 and S2 must be continuous across the interface). Perpendicular to the film plane, the thin film is usually free to move, i.e., T3=0. As a consequence, the complete equation of state needs to be analyzed. A convenient description for the parallel plate capacitor geometry is obtained when passing to mixed variable description with S1, S2T3. The direct effect is then written as:51




It turns out that d33,f is always smaller than d33 and that the absolute value of e31,f is always larger than that of e31. The dielectric constant is smaller than the one of a free body. To give an example: For the pure PZT ceramic,22 one calculates a reduction of the dielectric constant by 300 as compared with free (zero stress) value.

The situation with a ferroelectric equipped with interdigitated electrode is more complicated. Considering the electric field in the film plane along direction x (perpendicular to the electrode fingers) in the middle between interdigitated electrodes (running along y), we deal with a orthorhombic symmetry. The material constants along the poling direction x are different than along direction y, and both of them are different from the growth direction of the grains. The effective e-coefficients are derived from the following equation of state in case of clamping (the indices x, y, z are used to point out the difference to tetragonal or cylindrical symmetry):


A longitudinal exx,f and a transverse exy, f are defined. Evaluating these coefficients using some standard values for PZT as published by Berlincourt and Jaffe i.e., inline image, inline image, inline image, dxx (=d33)=260 pm/V, dxy (d31)=−90 pm/V yields a very high exx,f of 27 C/m2 and a rather small exy,f of 1.3 C/m2, which is by chance also positive. This means that there is almost a mono-axial bending, as required for cantilever deflection. The quality of such IDE structure depends of course on the success of the poling process requiring higher fields than with parallel plate geometry. However, there is much to gain, even if we consider some inactive zones below the electrodes fingers. The compressive piezoelectric stress may yield larger deflections of membranes (which was indeed observed52), and does not provoke cracking of PZT, as may occur in very thick, dense PZT thin films.

V. Ferroelectric Domains

In the previous section, the piezoelectric and dielectric coefficients were considered as constants. In ferroelectrics, though, nonlinearity is introduced by field-dependent coefficients.53 This is due to domain wall motions and switching. The ferroelectric phase of perovskites is derived from a cubic high-temperature phase. The ferroelectric phase is for instance tetragonally distorted as in Ti-rich PZT, giving rise to ferroelastic 90° domains, separated by domain walls in {110} planes. The lattice constant c along the polar axes becomes longer, the perpendicular two lattice constants equal to a become smaller. In PZT ceramics, about 40%–60%54,55 of the piezoelectric small signal response is due to ferroelastic domain wall motions. In an electric field, domains change volumes depending on whether their potential energy inline imageis increased or reduced. In the switched volumes, the direction of the spontaneous strains change, yielding a net strain change in the ferroelectric body. In thin films, ferroelastic domains adapt to strain or stress imposed by the substrate in order to reduce the elastic energy.56 In very thin films, such strains are due to epitaxial mismatch57,58; in thicker films, the cause is the thermal mismatch upon cooling down from growth temperature. Considering a {100}-oriented film, a domain pattern is formed compensating tensile stress by switching the c-axis into the plane (forming a a-domain), and compensating compressive stress by switching the c-axis out-of-plane (c-domain). The laminar c/a/c/a domain pattern is a possible solution to the elastic problem.59–61 The ratio of a- to c-domain volume fraction is given by substrate clamping and thermal mismatch history. However, the domain walls may still be free to move in an electric field compensating the increase in elastic energy. Analytical model calculations of the extrinsic piezoelectric effect (d33,f) in thin films are available for an ideal c/a/c/a domain pattern.60

During many years not much evidence of such domain contributions were found in thin films. It was thought that domain walls were very much pinned by defects. This idea was supported by hot poling experiments of tetragonal {100} PZT15/85 films.62 It was observed that hot poling helped to reduce the a-domain faction. This was attributed to unpinning from defects, mainly oxygen vacancies, because oxygen vacancies may move at the applied temperature of 150°–200°C. But after such poling, domains were even more fixed than before (which is advantageous for pyroelectric detectors). Later, it was observed that more than 3-μm-thick polycrystalline films deliver significant contributions from ferroelastic domain motions as indicated by considerable nonlinear effects in piezoelectricity.18,19

In recent years, much progress in hunting domain phenomena were made thanks to piezoelectric sensitive AFM measurements.63,64 Local piezoelectricity can be measured, and even individual domains can be observed. When patterning a 200-nm-thick PZT 40/60 film into dots with widths in the 100–200 nm range, thus smaller than the width, an increasing piezo-response with increasing aspect ratio was found65 (Fig. 9). As unclamping can account for at maximum a doubling of the response, the observed fourfold increase was attributed to unpinning of 90° domains and major domain reconfiguration taking place in addition to simple unclamping. Surprisingly such a drastic increase took place just by patterning i.e., without “additional help” such as for example annealing or poling, indicating that ferroelastic domain walls must be nevertheless quite mobile once the clamping is removed. Recently, a fivefold increase of the piezoresponse close to the etches of patterned features was observed when the film region was additionally subjected to poling before the measurement.66 In a very recent paper, it was reported that the response at a 90° domain wall boundary can be detected, even at a continuous film.67

Figure 9.

 The experimental elucidation of this problem was very much advanced by AFM investigations. A first indication of purely elastic clamping of ferroelastic domains was found at nanopatterned thin film structures with aspect ratio of up to 2. These were etched out from epitaxial PZT 40/60 films deposited on top of a conductive SrTiO3 crystal.65

VI. Piezoelectric Micromachined Structures in Deflective Mode

There is a plethora of applications using piezoelectric laminated structures in the bending mode using beam and plate structures. Rigidity and deflection amplitude can be engineered in a wide range. At the upper frequency range (0.1–15 MHz) we find piezoelectric micromachined ultrasonic transducers (pMUT), realized with ZnO68 and PZT.69,70 A pMUT is a device able to transmit and receive ultrasonic waves. Potential applications are ultrasonic imaging, nondestructive testing, droplet ejection,71 sensors using ultrasound, ultrasonic micromotors,72,73 and tonpilz transducers.33 A state-of-the-art version of integrated PZT films on micromachined structures is illustrated in Fig. 10(a). Dense and crack-free PZT films of 4 μm thickness can be deposited by sol–gel techniques on wafer level. The measured admittance curves (Fig. 10(b)) allow for an evaluation of the electro-mechanical coupling factor confirming very well the e31,f of 12–14 C/m2 or even more measured at beams.74–76 Such pMUT structures with strong PZT films are theoretically very much suited for ink jet printing heads. Resonators of flexural modes have also been proposed for analog signal processing, as needed for instance at intermediate frequencies (few MHz) in mobile communication.77,78 At lower frequencies, flexural structures were demonstrated for applications as microphones in photoacoustic gas sensors,79 as accelerometers,80 and more recently for energy scavenging.81,82 Similar to the accelerometer, the latter is combined with a seismic mass, and captures vibrations in a given frequency range. At an acceleration of 10 m/s (1 g) a generator resonating at 800 Hz yields about 100 μW/cm2.82 Important is an enough large voltage (i.e., more than about 200 mV) in order to be able to rectify the current, and accumulate charges by a charge pump onto a battery. The device cited above exhibits an output voltage of 1.8 V at 0.85 μW power (0.6 mm2 beam) thanks to interdigitated electrodes. It is thought that this kind of micropower generators are useful to feed wireless sensor nodes mounted at moving parts of machines, and serving to alert in case of machine problems. A further possible application is active damping of motion sensors. The same PZT film is used in the vibration sensor and the damping actuator.83,84 Actuation of optical mirrors is a further possibility.

Figure 10.

 Piezoelectric micromachined ultrasonic transducers: (a) SEM cross section of rectangular element. The elastic layer of the plate is formed by the device layer of a SOI wafer.76 (b) Admittance curve measured in air of circular element.75

There are much less works using the d33 perpendicular to the film plane. Displacements are of course small, but the response can be sufficient for sensor applications. Accelerometers85 and the sensor part in active damping have been proposed.84 Resonators in the thickness mode exhibit quite high frequencies in the lower GHz frequency region, where ordinary PZT has too high losses for applications. This application is very much suited for AlN.

VII. Bulk Acoustic Wave (BAW) Devices

Longitudinal BAW devices are known for their efficient transformation of electrical power into mechanical power and vice versa. A longitudinal ultrasound wave is trapped in an electroded, half-wavelength thick slab (thickness tp) of piezoelectric material having its polarization along the wave propagation (see Fig. 8(a)). In the bulk world this principle is very successfully applied in ultrasound probes for medical imaging and nondestructive testing at a few MHz. The antiresonance frequency inline image is inversely proportional to the thickness tp. With typical thin film thicknesses, thin film BAW resonators (TFBAR) resonate at frequencies of a few GHz. The high sound velocity vs of AlN of 11 000 m/s yields roughly 2 GHz at 2 μm (in reality somewhat less due to mass loading by electrodes). The typical film thickness range is thus ideal for RF filters in mobile communication (the US-CDMA standard requires for instance a duplex filter centered at frequencies 1960 and 1880 MHz for receive and transmit line, respectively). The elastic energy trapped in the TFBAR constitutes an energy tank of high quality as needed in RF signal processing. The electrical admittance of a TFBAR is the one of a capacitor (iwC0), modified near the resonance according to Ikeda86


where inline image is the thickness mode coupling factor, h is the thickness of the plate, and vD is the sound velocity at constant D-field. The losses enter through a complex sound velocity containing the complex stiffness (the imaginary part of the stiffness is proportional to the viscosity).

TFBARs are investigated since the early 1980s, initially mainly motivated by applications in television filters.87 The required frequencies were, however, too small, meaning that the required film thickness was too large to be feasible. The competing SAW devices were more easily produced at these low frequencies. In addition, sputter deposition was not yet the optimized technique with outstanding uniformity and reproducibility as it is today. The advent of mobile communication, the ever increasing process control in semiconductor technology, together with the momentum of MEMS research have lead to the convergence of all requirements for a new product called TFBAR RF filter. Today, it is one of the most successful MEMS products. The present market of around $100 million in 2005 is believed to extend to above $1 billion, reaching $300 million in 2008. BAW devices thus represent one of the most successful MEMS products. The basic element of a filter is a λ/2 plate resonator. There are two possibilities to achieve acoustic isolation of the film plate. In the first one a cavity is etched away underneath the resonator plate (Fig. 11(a)). This solution was historically the first one, and is often called TFBAR for thin film bulk acoustic wave resonator although this name does not refer to how the isolation is performed. Surface micromachining techniques are the preferred solutions to liberate the resonator.88 A sacrificial layer is deposited before the resonator layers, and removed afterwards. In a recent work with above IC integration, a polymer sacrificial layer was applied.89 A second possibility is the use of acoustic reflectors. A sequence of λ/4 layer pairs of low and high acoustic impedance layers very efficiently reflects the wave back into the resonator (Fig. 11(b)). This structure has been named solidly mounted resonator (SMR).90 This name reflects the fact that this structure is much more robust, and more suited for single component filter production,91 whereas for above-IC structures, the cavity solution appears to be the simpler one.

Figure 11.

 Schematic structure of bulk acoustic wave (BAW) resonators. (a) Free capacitor structure above micro machined cavity; (b) Solid-mounted resonator (SMR) structure with acoustic reflector providing acoustic isolation from substrate; (c) Stacked filter structure with SMR.

Resonators based on AlN thin films are able to operate at frequencies up to 10 GHz or even higher. Figure 12 shows the admittance of an AlN SMR resonator operating at 7.9 GHz showing still very decent characteristics concerning coupling and quality factor. In order to get a filter, several resonators are coupled together. Ladder filters and lattice filters are electrically connected resonators in which one type of resonator is shifted down so as to match its antiresonance frequency (zero susceptance) with the resonance frequency (peak of conductance) of the other type (see Lanz and Muralt92 for a description of the resonance shift for filter fabrication). A second possibility is acoustic coupling in stacked filters (see Fig. 11(c)). The signal passing from the first to the second resonator is an acoustic wave that is suitably transformed by intermediate matching layers.93

Figure 12.

 Admittance of bulk acoustic wave (BAW) solid mounted resonator (SMR) with AlN (001) having its fundamental resonance at 7.9 GHz.92

Currently, most of the BAW resonators are produced with AlN thin films. These are deposited by reactive sputtering in either RF94 or pluse DC mode.95 The alignment of the polar axis in all grains seems to be a result of ion bombardment combined with high atom diffusivity at the growing surface, as required for zone T growth.96,97 For this reason, sputter deposition appears to be the only method at lower temperatures (<400°C) able to produce good AlN thin films. More recently, the growth of c-axis tilted AlN was investigated.98,99 Large tilts of around 28° are obtained from grains nucleating in (103)-orientation.100

There are two key properties for the choice of the piezoelectric material in TFBAR's: the coupling coefficient inline image of the thickness mode and the material quality factor inline image. τ1 is a material constant. Both of these key properties are given for AlN and ZnO in Table I. It is interesting to note that the figure of merit for filter insertion loss, the product inline imageQm, is the same for both materials. As the bandwidth depends directly on inline image, ZnO looks more interesting at first sight. The fact is, however, that AlN is the favorite material at present. The coupling coefficient of AlN is sufficient to fulfill filter specifications. The decisive properties are a higher thermal conductivity and larger breakdown field, which allow for better power handling capabilities with AlN, enabling the use of AlN TFBARs in the transmit line. While there is hardly any material performing better than AlN at frequencies above 2 GHz, one may ask whether at lower frequencies, ferroelectrics could do the job. Ferroelectric TFBARs would be smaller and would add some tuning capabilities. Major problems are the acoustic losses due to domain wall relaxation (for a review, see, Muralt et al.101). Alternatively, electrostrictive materials like SrTiO3 or (Ba,Sr)TiO3 could be used, offering larger quality factors and no hysteresis because no domains are present in these materials.102 This is currently an intensively researched field. It is hoped to realize tunable filters and switching filters.

Table I.    Relevant Materials Parameters for TFBAR Applications for the Two Most Used Wurtzite Thin Films, as Derived from Standard Literature Data107
 inline image Z (%)τ1 (fs)Qm@ 2GHzinline imageQm@ 2GHzvs (m/s)
  1. TFBAR, thin film bulk acoustic wave resonators.

AlN6.532249016011 000

Resonating piezoelectric bodies are not only good for imaging and signal filtering. Piezoelectric resonators are also very sensitive to mass loading. This effect is used because decades for measuring the evaporation rate by means of quartz crystal oscillators, and also in may types of biological sensors.103 The relative sensitivity increases linearly with frequency. If we thus increase the frequency to the GHz range by means of a TFBAR, we could expect much more sensitive sensors than those operating at a few MHz! Recently, such sensors have been realized.104,105 The increase of sensitivity was verified, as shown in Fig. 13, which depicts the sensitivity reported in some selected works. The BAW numbers given in this figure correspond to longitudinal modes. Such devices could be useful as gas detectors. A thin polymer layer on the resonator would serve as gas sensitive layer changing its mass and its elastic constant with absorption of the gas. Detection in a liquid as required for biomedical sensors requires the use of shear mode BAW's. Longitudinal BAW's emit too much power (waves in liquids are only longitudinal) into the liquid causing excessive damping. Shear modes have no amplitude perpendicular to the film plane, but instead in the film plane. Hence, no propagating longitudinal wave into the liquid is excited, and a high Q is maintained in the liquid. The excitation of a shear wave requires an electric field that is inclined with respect to the polar axis of the film (S5=d15E1). This is realized either by the growth of polar films with tilted polar axis in combination with a usual planar capacitor,100,106 or the use of interdigitated electrodes.

Figure 13.

 Overview on sensitivity results obtained with quartz microbalances (QCM), surface acoustic wave (SAW) devices, and bulk acoustic wave (BAW) devices 104,105 (from ref Rey-Mermet et al.)105.

VIII. Conclusions and Outlook

The integration of piezoelectric materials in micro and nanosystems has made considerable progress during the last 10 years. The thin film material quality approaches, or even surpasses known bulk properties. Piezoelectric MEMS find increasing interest in applications for which piezoelectric actuation or transduction offers superior performance. The most relevant application to date is in analog signal processing for mobile communication. Bulk acoustic waves in AlN thin films are employed in RF filters. The pressure to reduce component number and size will push semiconductor industry to innovate on above-chip integration of more and more passives. Oscillators and medium frequency filters will follow. The GHz solution will be again a BAW device, but for lower frequencies Lamb wave structures and flexural resonators are more suited. A promising new material in the game is electrostrictive SrTiO3 enabling ultrasonic valve and tuning devices. BAW technology may also have an impact on chemical and biomedical sensors. A further application domain is in liquid delivery and droplet ejectors. Piezoelectrics offer a good stability over a wide temperature range, do not need electric fields across the liquid and can be operated at moderate voltages (PZT). There are many other applications under investigation like energy scavenging, active damping, mirror arrays and scanners, and further more. Especially for PZT, there is still some knowledge gap in reliability issues, such as life-time stability in various environments. PZT thin films reached the properties of undoped ceramics. After having solved electrode stability, texture control, and homogeneity issues, it is probably time to advance with doping of PZT thin films, as this is the most obvious difference to superior ceramics. Of particular interest is also domain engineering. The question is whether certain domain configurations are more favorable for a high response than others, and how they can be established by suitable growth conditions, electrode geometries, and electrical treatments.

inline image Professor Paul Muralt is group leader for thin film and MEMS activities at the Ceramics Laboratory of Swiss Federal Institute of Technology EPFL at Lausanne, Switzerland. He received a diploma in experimental physics in 1978 at the Swiss Federal Institute of Technology ETH in Zurich, and accomplished his Ph.D. thesis at the Solid State Laboratory of ETH. In the years 1984 and 1985 he held a post doctoral position at the IBM Research Laboratory in Zurich where he pioneered the application of scanning tunneling microscopy to surface potential imaging. After a stay at the Free University of Berlin, he joined the Balzers group in Liechtenstein in 1987. He specialized in sputter deposition techniques, and managed since 1991 a department for development and applications of Physical Vapor Deposition processes. In 1993, he joined the Ceramics Laboratory of the Swiss Federal Institute of Technology EPFL in Lausanne, where he started activities in ferroelectric thin films and MEMS devices. His interests are in thin film growth and integration issues of ferroelectric and other polar materials, property-microstructure relationships, and applications of polar materials in semiconductor and micro-electro-mechanical devices. More recent work deals with the fabrication and property assessment of small ferroelectric structures, and furthermore oxygen ion conductors. As teacher, he gives lectures in thin film deposition, micro patterning, and ceramics. He authored or co-authored more than 200 scientific articles.