Review of energy harvesting techniques and applications for microelectronics

Review of Energy Harvesting Techniques and Applications for Microelectronics Loreto Mateu and Francesc Moll Universitat Polit`ecnica de Catalunya Dept. of Electronic Engineering Barcelona, Spain ABSTRACT The trends in technology allow the decrease in both size and power consumption of complex digital systems. This decrease in size and power gives rise to new paradigms of computing and use of electronics, with many small devices working collaboratively or at least with strong communication capabilities. Examples of these new paradigms are wearable devices and wireless sensor networks. Currently, these devices are powered by batteries. However, batteries present several disadvantages: the need to either replace or recharge them periodically and their big size and weight compared to high technology electronics. One possibility to overcome these power limitations is to extract (harvest) energy from the environment to either recharge a battery, or even to directly power the electronic device. This paper presents several methodstodesignanenergyharvesting device depending onthetypeofenergyavaliable. Keywords: Energy harvesting, wearable devices, wireless sensor networks 1. INTRODUCTION One of the most important trends in the electronic equipment technology from its origins has been the reduction in size and the increase in functionality. Nowadays small, handheld, though very powerful devices are commercially available that allow the user to play music, to wirelessly communicate or to compute practically everywhere or, in other words, ubiquitously. In the next years there will be new products available providing vision and other extended functions to the wearer. The size of such devices is becoming so small that instead of portable devices they are becoming wearable devices that can be integrated in everyday use objects like watches, glasses, clothes, etc.1 All those units, based on today’s microelectronic technology, need an external power supply. The size of the electronic circuit and the energy needed to perform a single (binary) operation2has been drastically reduced during the last decades, following Moore’s Law. Therefore, advances in low power design open the possibility to harvest energy from the environment to power electronic circuits. This work presents current approaches of using energy extracted from the environment to power microelectronic devices. The electrical energy to power the electronics is generated from either kinetic, electromagnetic or thermal energy. The obtained energy can then be used to recharge a secondary battery or, in some cases, to power directly the electronics. The output voltage and current of the generators is transient and discontinuous in nature, and must be converted to a DC signal. Therefore it is necessary to design a converter and/or storage circuit that needs to take into account the output signal of the generator and its impedance. The structure of the paper is as follows: First, in section 2, the technology trends for both battery (energy generation) and electronic devices (energy consumption) are briefly sketched. Then, energy harvesting devices areclassifiedinsection3accordingtosourceandtypeof energy. Sections 4 through 6 review the principles behind different types of energy conversion. In section 7 it is discussed the power conditioning needed by electronic devices relying on energy harvesting, depending on their consumption and energy generation possibilities. Finally, in section 8, the conclusions of the work are presented. Further author information: (Send correspondence to F. Moll) F. Moll: E-mail: [email protected]c.edu L. Mateu: E-mail: mloret[email protected]c.edu Table 1. Characteristics of batteries. Battery type Vol. Energy density Grav. Energy density Self-discharge Cycle Life Wh/dm3Wh/kg %peryear no. Alkaline 300 125 4% 1 Ni-Cd 100 30-35 15-20% 300 Ni-MH 175 50 20% 300 Li-ion 200 90 5-10% 500 2. TECHNOLOGY TRENDS 2.1. Battery evolution In the great majority of today’s wearable or portable devices, the energy necessary for their operation is provided by batteries. Batteries are a significant fraction of the total size and weight of the system. As technology scales down, this fraction is expected to further increase. Also very important is the requirement for proper maintenance of batteries, with the need to either replace or recharge them. This is a serious limitation to computing paradigms like ubiquitous computing or sensor networks, in which there are dozens or hundreds of small systems with batteries to maintain. Of course, these inconvenients do not hide the advantages of batteries as a reservoir energy source. For example, we can characterize the batteries by their energy density, with respect to volume and weight, called volumetric and gravimetric energy density respectively. Table 1 shows some typical values of energy densities and self-discharge values for commercial batteries. It is seen how the most modern batteries (Li-ion) present better characteristics of energy density, self-discharge current and number of cycles. It is worth noting that these values of energy density are the best option available today. Recent advances in capacitor technology have led to the development of the so-called ultracapacitors, with a capacitance value of the order of kF. Such impressively large capacitors, however, present an energy density around 3 Wh/kg, very far from average battery values.3The main advantages of ultracapacitors is the supplied peak power, and the number of cycles. These characteristics make them more oriented to automotive applications than to low power electronic devices, where batteries are still the choice for energy storage. Still, battery technology has evolved very slowly compared to electronic technology.4For example, while disk storage density has increased over 1,200 times since 1990, battery’s energy density has increased only about 3 times. New approaches are on the way for very small size portable batteries that may enable the size and weight reduction of wearable systems and sensor networks. They are based on new technologies as the thinfilm Lithium-ion or Lithium polymer cells and some of them are still under development.5One of the most promising alternatives is the fuel cell, power generators that use chemical fuels (i.e. Hydrogen or Methanol). The gravimetric energy density of fuel cells is expected to be three to five times larger than Li-ion cells and more than ten times better than Ni-Cd or Ni-MH batteries whereas the volumetric energy density is six to seven times larger than Li-ion.6However, the maintenance problem is not solved; these cells need to be refueled or alternatively they have to be manufactured carrying enough fuel to sustain the battery operation during all its expected lifetime. Secondary batteries are in principle a better choice for ubiquitous or wearable systems because they can be recharged in several ways, in many cases without extracting the battery from the system. Actually, one of the possibilities to recharge such batteries is to use energy harvested from the environment.7In this sense, energy harvesting is not trying to replace batteries, but instead alleviating some of their drawbacks, especially in relation with the maintenance issue. 2.2. Power consumption of microelectronic devices Following Moore’s law, integrated circuit technology evolves following a transistor size shrinking trend. Together with this trend and because of reliability reasons the supply voltage (VDD) is also reduced. The net result is a reduction in energy consumption due to the reduction in the size of parasitic components. For a scale reduction Table 2. Classes of energy harvesting devices. Energy Source Type of Energy Human Kinetic, Thermal Environment Kinetic, Thermal, Radiation with a factor α(α>1), the energy consumed by a given shrunk circuit performing a given task is reduced by (1/α)3, as dicussed elsewhere.8 It is also important to consider the use model of the electronic device. For example, the power consumption of a wearable can be scaled following two different scenarios: 1. Maximum performance use: the improvement in technology allows to reduce the time per service, but it is used to give a higher number of services. In this case the total power consumption for the processing functionsisscaledas(1/α)2. 2. Constant number of services: the improvement in technology reduces the time per service and the power consumption but the user does not increase the number of required services. The power consumption for the processing functions is scaled as (1/α)3. Current portable electronic devices have different low power or sleep modes to save energy during times of inactivity. The management of these modes is very important in relation with an energy harvesting strategy, allowing to “refill” the energy reservoir of the system during these periods of low activity. This means that generally, a discontinuous operation use model is mandatory for the energy harvesting approach. 3. ENERGY HARVESTING DEVICES Before starting to discuss the different methods of energy harvesting, let us look at the definition of an energy harvesting device as understood in this paper. An energy harvesting device generates electric energy from its surroundings using some method of what is called in the literature Direct Energy Conversion techniques.9Therefore, the energy harvesting devices here considered do not consume any fuel or substance, so that the maintenance problem mentioned above does not apply. On the other hand, as the environment energy levels are very low (at least for today’s electronic devices requirements), the use models mentioned in the previous section are very important. 3.1. Classification of Energy Harvesting devices We may classify the different energy harvesting devices in two ways: considering who or what provides the energy for conversion, and what type of energy is converted. Table 2 relates the two classification schemes. In the first classification scheme we can distinguish between two kinds of devices. First, devices that use part of the energy of the user of the electronic appliance. It will usually be a human, but it could be also an animal, for example for a remote monitoring device. We call this first kind of devices Human Energy∗devices.10 The second kind of energy harvesting device gets its energy from the environment, and thus we call them Environment Energy devices. This classification takes into account that, following the first principle of thermodynamics, a greater amount of energy must be spent to obtain a certain amount of electrical energy. In the case of Human Power it is the user that in some way or other provides this energy and, though the energy levels are very small, the effect may prove noticeable when several devices depend on the activity of a single user. In order to evaluate the burden of energy harvesting on user activities, it is possible to use a simple biomechanical model to calculate the energy involved in a human step,11 obtaining around 40 J. In comparison, the energy of a short RF transmission can be evaluated in the order of 100 µW. This means that the extra energy demanded to obtain ∗In most of the literature the term Power is used instead of Energy. We prefer the term Human Energy instead of Human Power because energy is a more meaningful magnitude when dealing with discontinuous events, as discussed in subsection 7.1. enough energy is very small for some applications, and therefore it makes sense to consider human beings as a possible energy source. The second classification scheme may consider three types of energy: kinetic, electromagnetic radiation (including light and RF), and thermal. For Human Energy devices only kinetic and thermal energy are available. In the case of kinetic Human energy, one may distinguish between those actions made specifically to generate energy and casual movements made during normal behavior. These two cases are called by the Human Power research group of the Delft University of Technology12 Active and Passive Human Energy respectively. Following this definition, thermal Human Energy is always passive. Environment Energy sources include kinetic energy in the form of vibrations, radiation as solar energy or RF radiation, and thermal energy. The energy harvesting devices may pick up vibrations when located on machines, building elements or other places near vibrating sources. Radiation may come from natural or artificial sources. Thermal energy depends on the existence of a temperature gradient. While the transducing methods may be similar to the Human Energy devices, the excitation magnitudes, frequency spectra and periodicity are very different, and therefore each case must be studied separately. This will also have consequences in the electrical power conditioning circuit. 3.2. Some working examples There are several examples, both commercial and in the research stage that apply energy harvesting to power electronic products. 3.2.1. Human Active Energy This is an old concept that has revived recently, basically improving the ratio of time of use with respect to time of charge. Devices in the market using human energy as the only energy source include radio receivers, electric torches and phone battery chargers.13–15 These devices use kinetic energy provided by winding a hand crank, or shaking the device, and they offer a good ratio between charging time and use time.16 3.2.2. Human Passive Energy While Human active energy is interesting as an industrial concept, it is Human Passive Energy that presents a real challenge and is most attractive because it eliminates the power maintenance problem in portable and wearable devices. Among commercial products, the first devices were wristwatches, because they have a very low power consumption. Both kinetic and thermal energy powered watches have been commercialized, although currently, only the kinetic is being manufactured by several companies. For kinetic energy, the power output is 5 µWinnormal conditions, and up to 1 mW when the watch is forcibly shaken. For thermal conversion, around 1.5 µWormore is generated when the temperature difference is 1–3◦C.17 Miniature thermoelectric generators are also developed18 that converts body heat flow into electricity. It is claimed that it can generate 40 µW at 3 V with a 5 degree difference in temperature. Potential applications include attachable medical devices, electronic wrist watches, self powered heat sensors, and mobile electronics. Another application already commecialized for energy harvesting is in the remote control area. The mechanical force employed to push down the switch is used to bend a cantilever piezoelectric ceramic.19 This kind of devices is considered as Human Passive energy harvesting because the same movement of pushing the switch is used to generate the energy for a wireless communication. 3.2.3. Environment Energy Several approaches have been made during the last few years in order to harvest energy from the environment to power wireless sensor networks. Batteries are not a recommended power source for wireless sensors since the power source would limit the lifetime of the sensor. The energy needed by a wireless sensor is in the order of hundreds of micro watts. The main power sources studied for wireless sensor networks are solar power (outdoors or indoors) and mechanical vibration. Roundy et al.20 analyzed and fabricated a bimorph piezoelectric (PZT) generator with a steel center shim. The cantilever structure has an attached mass. The volume of the total structure is 1 cm3. A model of the developed piezoelectric generator was made and validated. For an input vibration of 2.25 m/s2at about 120 Hz, power from 125 µW to 975 µW were generating depending of the load. The power recovered was analyzed connecting the generator directly to a resistive load or to a capacitive load. Later, a DC-DC converter was included and the generator supplied power to a low power transceiver. The radio transmits at 1.9 GHz and consumes 10 mA at 1.2 V and the vibration source was 2.25 m/s2at 60 Hz. 4. KINETIC ENERGY Kinetic energy is one of the most readily available energy source, both for Human and for Environment energy harvesting devices. This section briefly explains the principles of the different transducers for obtaining electrical energy from kinetic energy. 4.1. Types of kinetic energy transducers The principle behind kinetic energy harvesting is the displacement of a moving part or the mechanical deformation of some structure inside the energy harvesting device. This displacement or deformation can be converted to electrical energy by three methods, that are explained in subsequent subsections: by a piezoelectric material (subsection 4.2), by electrostatic energy (subsection 4.3) and by magnetic induction (subsection 4.4). With respect to mechanical structures, there are two types of possible converters. One responds to the kinetic energy with a vibration or displacement of a proof mass. The energy obtained will depend on this mass, and therefore we will call this first class Inertial converters. Mitcheson et al.21 have classified inertial converters in function of the force opposing the displacement of the proof mass as Voltage Damped Resonant Generators (VDRG), appropriate to describe magnetic induction transducers, Coulomb Damped Resonant Generators (CDRG), which describe vibrating electrostatic transducers, and Coulomb Force Parametric Generators (CFPG) that correspond to displacement electrostatic generator type. There are also inertial converters based on piezoelectric, in which an accelerated mass causes a deformation of a piezoelectric material, either by impact or vibration. Many inertial converters are based on a spring-mass system that resonates at a particular frequency. When the mechanical stimulus vibrates at that resonance frequency, the enery obtained is maximum. However, as the converters are miniaturized to integrate them on microelectronic devices, the resonance frequency increases, and it becomes much higher than characteristic frequencies of many everyday mechanical stimuli. For example, typical acceleration frequencies of the human body in movement are below 20 Hz.22 As was recognized by Mitcheson et al.21 for such cases, either a CFPG type of converter, or a non inertial converter is best suited. In the second case (Non-Inertial converters), an external element applies pressure that is transformed as elastic energy, causing a deformation that is converted to electrical energy by a piezoelectric material. In this case, the obtained energy does not depend on the mass of the converter, but generally, on the rate of deformation, thus given by mechanical constraints like Young’s modulus or geometric dimensions.23 4.2. Piezoelectric generator The piezoelectric effect was discovered by Jacques and Pierre Curie in 1880. Curie’s brothers found that certain materials, when subjected to mechanical strain, suffered an electrical polarization that was proportional to the applied strain. This is the piezoelectric effect used for mechanical to electrical energy conversion. The phenomenon of piezoelectricity is described by the following equations: {Si}={1 Yc,ij }{Tj}+{dik}{Ek} {Dl}={lm}{Em}+{dln}{Tn}for j, n =1,...,6andi, k, l, m =1,2,3(1) Table 3. Subscripts of the reduced notation for piezoelectric constitutive equations. Reduced notation Corresponding direction of axes 1Longitudinal in xdirection 2Longitudinal in ydirection 3Longitudinal in zdirection 4Shear y-z 5Shear z-x 6Shear x-y C'p T 1:-1/(dYc) SD E 1/Yc C'm S T 1:-d/ε D Cp E (a) (b) Figure 1. Piezoelectric coupling circuits, relating mechanical and electrical magnitudes. In these equations, subscripts correspond to the 6 directions of the axes, three cartesian directions plus the shear around the three axes, as shown in Table 3. Repeated subscripts in the products imply a summation over the different components: T, applied mechanical stress [N/m2]. E, applied electric field [N/C]. d, piezo strain tensor [(C/m2)/(N/m2)]. , permittivity tensor [F/m]. D, electric displacement [C/m2]. S, mechanical strain [m/m]. Yc, Young’s modulus tensor [N/m2]. A piezoelectric material mechanically stressed at a low frequency can be modelled electrically by a timedependent charge source, that is accumulated in a capacitor. If the piezoelectic constitutive equations are transformed to the Laplace domain, the following equations are obtained: ˙ S=jω T Yc+jωdE ˙ D=jωTE+jωdT (2) This relationship is represented by the piezoelectric coupling circuit of Figure 1(a). The transformer relates mechanical magnitudes (stress) and electrical magnitudes (electric field). C p=T1−d2Yc T=T(1 −k2)(3) Another possibility for the representation of the piezoelectric coupling circuit is shown in Figure 1(b), where capacitors C mand Cpare given by: C m=1 Yc1−d2Yc =1 Yc(1 −k2);Cp=TA t(4) The two most common types of piezoelectric materials are PVDF, polyvinylidene fluoride, and PZT, lead zirconate titanate. There are three different ways to excite a piezoelectric material in order to generate electrical energy: by compression, slap and bending. For the analysis of the piezoelectric response, it is assumed that electrical terminals are located along direction 3, parallel to the poling axis (the direction of the polar molecules that form the piezoelectric effect). The two most common ways to employ piezoelectric materials are modes 31 and 33. In mode 31, the stress is applied in Table 4. Voltage V3and charge q3obtained applying a mechanical stress in direction 1, mode 31, and in direction 3, mode 33. Constants gij are defined as dij /. Mode 31 Mode 33 V3g31 F1 Wg33 F3 WLH q3d31 F1L Hd33F3 direction 1 whereas the electric field (voltage mode) or the electric displacement (charge mode) are in direction 3. In mode 33, the stress is applied in direction 3. Table 3 gives the obtained voltage or charge resulting from an applied force in a certain direction, F1or F3 for each mode. If the same force is applied in direction 3 and in direction 1 over a piezoelectric material with similar dimensions in length L,widthWand thickness H, mode 33 excitation can generate more charge and voltage than mode 31 because d33 is usually larger than d31. However, the geometrical dimensions of the material play a very significant role. For example, in a thin PVDF film, the ratio L/H is on the order of 1000, while d31 =23·10−12m/V and d33 =−33 ·10−12m/V .24 If it is considered again that F1is equal to F3,theobtained value for V3and q3for the mode 31 will be on the order of 700 times greater than for the mode 33. Therefore, the mechanical structure must take into account both the geometry and the mechanical coupling in order to apply the deformation in the optimum way. Piezoelectric thin films are used for several systems due to their adaptability. In this case, the most advantageous way of excitation is by bending piezoelectric materials configured in cantilever-like structures. There are different kinds of support, and different vertical structures of the material that is going to be bent. The combination of these structures generates several options.23 When several piezoelectric elements are present in the structure, they can be connected either in parallel or in series. In a parallel connection, the charge generated by the piezoelectrics is added whereas in a series connection, the charge generated corresponds to the strain of one of the piezoelectric elements connected, and the voltage of the piezoelectric elements is added. In order to adequately connect the piezoelectric elements in series or parallel, the orientation of the poling axis has to be taken into account. In summary, piezoelectric converters have most of the advantages of inductive and electrostatic generators, and they are also very robust. On the other hand, piezoelectric converters are difficult to implement on micromachined processes, and therefore to miniaturize.20 4.3. Electrostatic energy generator The principle of electrostatic generators is that the moving part of the transducer moves against an electrical field, thus generating energy. Meninger et al.25 of MIT presented an electrostatic generator that employs a variable micromachined capacitor. Two different designs were studied: a parallel capacitor operated with a constant charge and a comb capacitor operated with a constant voltage. These generators are also called Coulomb-damped resonant generators (CDRGs) because they are based on electrostatic damping. If the charge on the capacitor is maintained constant while the capacitance decreases (e.g. reducing the overlap area of the plates or increasing the distance between them), the voltage will increase. If the voltage on the capacitor is maintained constant while the capacitance decreases, the charge will decrease. Figure 2 illustrates the process of charging and discharging the capacitance following constant charge (path A-B-D-A) or constant voltage (path A-C-D-A) approaches. The energy enclosed by the total path is the energy extracted in the process. The charge constrained conversion cycle starts when the micromachined capacitance (given by the slope of the Q-V curve) is maximum. At this moment, a voltage source charges the MEMS capacitor to an initial voltage, Vstart, that has a smaller value than Vmax, and therefore the cycle conversion goes from point A to point B. The path B-D corresponds to the plates moving from maximum capacitance, Cmax, to minimum capacitance, Cmin cmax cmin Voltage-constrained cycle Charge-constrained cycle V Q Vstart Vmax A B C D Qa Figure 2. Diagram explaining electrostatic energy conversion. Adapted from Meninger et al.25 with constant charge, Q0. As the capacitor decreases and charge is maintained constant, the voltage increases its value. The charge is returned to the reservoir in path D-A. The net energy out is equal to the area A-B-D. The voltage constrained conversion cycle starts when the micromachined capacitance is maximum. At this moment, a voltage source charges the MEMS capacitor to an initial voltage, Vmax, and therefore the cycle moves from point A to point C. The path C-D corresponds to the plates moving from maximum capacitance, Cmax, to minimum capacitance, Cmin. Path D-A shows the discharge of the capacitor. The mechanical vibration that takes place in path C-D is converted to electrical energy with a constant voltage. The net energy gained corresponds to the area A-C-D. The energy gained in the conversion process is pumped from the MEMS capacitor along path DA for both charge and voltage constrained cycle. As shown graphically, the mechanical energy converted into electrical energy is greater if the voltage across the capacitor is constrained than if the charge across the capacitor is constrained. However, the initial voltage source needed has a smaller value for the constant charge case. A way to increase the electrical energy for the charge constrained method is to add a capacitor in parallel, Cpar with the MEMS capacitor, CMEMS. The disadvantage of this solution is that the initial voltage source has to increase its value. As explained, the proper operation of the switches, or when the charges are transferred, is critical for a good efficiency. The operation of the switches must be synchronized with the mechanical oscillation. The frequency of the mechanical oscillation depends on the resonance frequency of the mechanical structure. For best results, then, the mechanical source must have a vibration with a frequency close to that of the resonance frequency of the transducer, which is in the order of kHz for miniaturized components. Other approaches use non-resonant structures21 that are more suited for mechanical excitations at lower frequencies. 4.4. Magnetic induction generator The magnetic induction transducer is based on Faraday’s law. The variation in magnetic flux, Φmthrough an electrical circuit causes an electric field. This flux variation can be realized with a moving magnet whose flux is linked with a fixed coil or with a fixed magnet whose flux is linked with a moving coil. The first configuration is preferred to the second one because the electrical wires are fixed. As the relevant magnitude here is the magnetic flux through a circuit, the size of the coil is inversely related to the obtained electric field and therefore, to the generated energy. This means that big transducers with large m k Bm y(t) z(t) R L Rc - + fm V Z ksBms 1 m 2++ c RRLs BlRs ++ R Bl fe (a) (b) Figure 3. Magnetic induction transducer model. Adapted from Ching et al.26 area coils will perform better than smaller transducers, unless a larger time derivative is involved with the small scale generators. We briefly present the analysis of a simple generator.26 When the generator vibrates, the oscillating mass has a relative displacement with respect to the housing. The magnetic induction generator converts this relative displacement into electrical energy. The transducer is modelled as a damped spring-mass system, since the energy extraction damps the mass movement with a factor Bm.Themassm,themagnet,whichisjoinedtoa spring with a spring constant kmoves through a constant magnetic field, B, when the generator oscillates. The relative displacement, z(t), is related to the voltage across the coil by a first order system as can be shown in Figure 3(a). Lis the inductance of the coil, Rcis the parasitic resistance of the coil, lis the length of the coil, and Ris the load resistance. Figure 3(b) shows the transfer function block diagram that relates the mechanical input force with the output voltage. The current induced in the coil generates an electromechanical force, fethat dampes the movement of the magnet. From the Newton’s second Law of motion, the transfer function between the input mechanical force, fm, and the relative displacement of the mass can be obtained. After some algebra, the transfer function that relates the output voltage V, across the load resistor with the input mechanical force fmis: V(s) fm(s)=(BlR)s (Ls +R+Rc)(ms2+Bms+k)+(Bl)2s(5) The third order system can be simplified to a second order system by assuming that the electrical time constant is much smaller than the mechanical time constant. With this assumption, the transfer function is expressed as: V(s) fm(s)= (Bl)s m (s2+2ζωns+ω2 n)(6) where ζand ωnare the damping factor and the spring natural frequency, respectively. ωn=k m(7) ζ=BmR+(Bl)2 2R√mk =Bm 2ωnm+(Bl)2/R 2ωnm=ζm+ζe(8)