IEEE Wireless Communications • December 201018 1536-1284/10/$25.00 © 2010 IEEE

THE IN T E R N E T O F TH I N G S

INTRODUCTIONThis article focuses on the design challengesposed by a new class of ultra-low-power devices— Energy-Harvesting Active Networked Tags(EnHANTs). EnHANTs are small, flexible, andself-reliant (in terms of energy) devices that canbe attached to objects that are traditionally notnetworked. The realization of EnHANTs isbased on recent advances in the areas of solar

and piezoelectric energy harvesting [2] and inparticular in the area of organic semiconductorsthat allow having flexible solar cells [3], therebyenabling pervasive use of tags. Another enableris the development of short-range ultra-low-power wireless communications circuits, in par-ticular for ultra-wideband (UWB)communications [4].

Following the transition from barcodes toRFIDs, we envision a future transition fromRFIDs to EnHANTs that:• Network: Actively communicate with one

another and with EnHANT-friendly devices inorder to forward information over a multihopnetwork.

• Operate at ultra-low-power: Spend a few nano-Joules or less on every communicated bit.

• Harvest environmental energy: Collect andstore energy from sources such as light,motion, and temperature gradients.

• Are energy adaptive: Alter communicationsand networking to satisfy energy and harvest-ing constraints.

• Exchange small messages: Exchange limitedinformation (basically IDs) using low datarates.

• Transmit to short ranges: Communicate onlywhen in close proximity (1 to 10 meters) toone another.

• Are thin, flexible, and small (a few square cen-timeters at most).As shown in Fig. 1, in terms of complexity,

throughput, size, and energy requirements,EnHANTs fit between RFIDs and sensor net-works. Similarly to RFIDs, the tags can be affixedto commonplace objects. Presence of powersources and distributed multi-hop operation shiftEnHANTs closer to sensor networks than toRFIDs that are traditionally passive and not net-worked. Compared to sensor nodes, however,EnHANTs operate at significantly lower datarates, consume less energy, and transmit mostlyID information.

EnHANTs will be enablers for the Internet ofThings and as such will support a variety oftracking and monitoring applications beyond

MARIA GORLATOVA, PETER KINGET, IOANNIS KYMISSIS, DAN RUBENSTEIN, XIAODONG WANG,AND GIL ZUSSMAN, COLUMBIA UNIVERSITY

ABSTRACTThis article presents the design challenges

posed by a new class of ultra-low-power devicesreferred to as Energy-Harvesting Active Net-worked Tags (EnHANTs). EnHANTs are small,flexible, and self-reliant (in terms of energy)devices that can be attached to objects that aretraditionally not networked (e.g., books, furni-ture, walls, doors, toys, keys, produce, and cloth-ing). EnHANTs will enable the Internet ofThings by providing the infrastructure for vari-ous novel tracking applications. Examples ofthese applications include locating misplaceditems, continuous monitoring of objects, anddetermining locations of disaster survivors.Recent advances in ultra-low-power circuitdesign, ultra-wideband (UWB) wireless commu-nications, and organic energy harvesting tech-niques will enable the realization of EnHANTsin the near future. The harvesting componentsand the ultra-low-power physical layer have spe-cial characteristics whose implications on thehigher layers have yet to be studied (e.g., whenusing UWB communications, the energyrequired to receive a bit is significantly higherthan the energy required to transmit a bit). Inthis article, we describe paradigm shifts associat-ed with technologies that enable EnHANTs anddemonstrate their implications on higher-layerprotocols. Moreover, we describe some of thecomponents we have designed for EnHANTs.Finally, we briefly discuss our indoor light mea-surements and their implications on the designof higher-layer protocols.

ENERGY HARVESTING ACTIVENETWORKED TAGS (ENHANTS) FORUBIQUITOUS OBJECT NETWORKING

The authors describeparadigm shifts associated with technologies thatenable EnHANTs anddemonstrate theirimplications on higher-layer protocols.

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 18

IEEE Wireless Communications • December 2010 19

what RFIDs permit. While RFIDs make it possi-ble to identify an object in proximity to a reader,EnHANTs will make it possible to search for anobject in a network of devices, and to continu-ously track objects’ whereabouts and their prox-imity to each other. Unlike RFIDs that areactivated when placed near a reader, and reportonly on themselves, EnHANTs can operate con-tinuously, can achieve pervasive coverage due totheir networking capabilities, and can report onthemselves and other EnHANTs around them.

One application that we plan to demonstratein the near future is a misplaced library booklocator.1 The initial prototype will enable librarybooks to identify those among themselves thatare significantly misplaced (e.g., in an incorrectsection), and report the misplacement. Toaccomplish this task, each book is assigned aunique ID using an assignment scheme closelyrelated to the Dewey Decimal Classification.Each book has a solar powered tag whose poweroutput is sufficient to transmit and receive infor-mation within a radius of one meter or less, andto perform some basic processing. Nearby bookswirelessly exchange IDs, and IDs of books thatappear out of place are further forwardedthrough the network of books, eventually propa-gating to sink nodes. A long term objective couldbe to determine book locations when books arearbitrarily ordered and to propagate that infor-mation to a central server using the multihopwireless network.

The same building blocks used in the libraryapplication can enable several other applica-tions. A large variety of items can be tracked anda range of possible desirable or undesirable con-figurations of objects can be queried for, and cantrigger reports. Examples include finding itemswith particular characteristics in a store, continu-ous peer monitoring of merchandise in transit,locating misplaced items (e.g., keys or eyeglass-es), and locating survivors in disasters such asstructural collapse.

EnHANTs are brought about by the recentadvances in ultra-low-power communicationsand energy harvesting technologies. However,current RF transceiver designs, communicationand networking protocols, and energy harvestingand management techniques have been devel-oped in isolation and are inadequate for theenvisioned EnHANTs applications. Althoughnetworking protocols for energy harvestingnodes recently gained attention (e.g., [6–8]), thecross layer interactions between circuit design,energy harvesting, communications, and network-ing have not been studied in depth. Hence, in this

article, we outline the cross-layer design chal-lenges that are posed by this new technology,and describe some of the components we havedeveloped. Moreover, we briefly provide resultsof our indoor light measurement study and dis-cuss their implications on the design of proto-cols.

This article is organized as follows. First, wedescribe related work. Then we discuss energyharvesting and energy storage techniques, as wellas ultra-low-power EnHANT communications.We also discuss EnHANT networking as a large-scale optimization problem. Finally, we brieflydescribe an EnHANTs testbed and presentssome results of our indoor light measurements.

RELATED WORK

While the idea of pervasive networks of objectshas been proposed before, the harvesting andcommunications technologies have recentlyreached a point where networked energeticallyself-reliant tags are becoming feasible.EnHANTs are likely to be implemented by com-bining flexible electronics technologies (a.k.a.organic electronics) with CMOS chips support-ing Impulse-Radio UWB. By using Impulse-Radio UWB, data is encoded by very shortpulses (at the order of nano-seconds) and theenergy consumption is very low. Several MACprotocols for IR-UWB sensor networks havebeen proposed, including the IEEE 802.15.4astandard which inherits many of the IEEE802.15.4 (ZigBee) functionalities. However, theoverhead to provide ranging, high data rates (upto 26Mb/s), and backward compatibility will leadto energy consumption largely exceeding theenergy anticipated to be available to EnHANTs.

Energy efficiency in wireless networks haslong been a subject of research. Energy-efficientrouting, for example, can serve as a basis forEnHANTs routing. However, energy harvestingnetworks have special characteristics and havebeen considered only recently. A number ofsolar energy-harvesting based systems have beenimplemented and deployed (e.g., [9] and refer-ences therein). Current deployments are mostlyfocused on outdoor energy and use much moreenergy than EnHANTs will have available. Forenergy-harvesting sensor networks, adaptive dutycycling [6], dynamic activation of wireless sensors[7], and adaptive data rate control [8] are amongthe exciting research directions.

Numerous characterizations and measure-ments of outdoor light energy are available.However, to the best of our knowledge, mostindoor light measurements have focused on aes-thetic and ergonomic attributes of light, ratherthan on its energy aspects. Indoor light is exam-ined from an energy perspective in [10]. Yet,there is insufficient data regarding the character-istics of energy available to EnHANTs in indoorsettings.

ENERGY HARVESTING

In this section we briefly describe two major har-vesting methods and possible energy storagemethods, as well as energy-harvesting and stor-age devices we developed.

Figure 1. EnHANTs in comparison to Sensor Net-works and RFIDs.

Throughput C

omplexity

Size Energy consum

ption

Sensor networks

EnHANTs

RFIDs

While the idea ofpervasive networksof objects has been

proposed before, theharvesting and

communicationstechnologies haverecently reached a

point where networked

energetically self-reliant tags arebecoming feasible.

1 An RFID-based solution(involving a mobile read-er) to the problem oflocating misplaced librarybooks has been proposedin [5].

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 19

IEEE Wireless Communications • December 201020

ENERGY SOURCES

Many environmental sources of energy arepotentially available for harvesting by smalldevices, i.e., temperature differences, electro-magnetic energy, airflow, and vibrations [2].Here we focus on the most promising harvestingtechnologies for EnHANTs: solar energy andpiezoelectric (motion) harvesting.

Solar Energy Harvesting — Solar energy (light) isone of the most abundant energy sources, withtypical irradiance (total energy projected andavailable for collection) ranging from100mW/cm2 in direct sunlight to 0.1mW/cm2 inbrightly lit indoor environments (notice the sig-nificant difference) [2, 10]. The efficiency of asolar energy harvesting device is defined as thepercentage of the available energy that is actual-ly harvested. Conventional single crystal andpolycrystalline solar cells, such as those com-monly used in calculators, have efficiencies ofaround 10–20 percent in direct sunlight. Howev-er, their efficiency declines with a reduction inenergy availability (they are less efficient withdimmer sources), which is important to note dueto considerable irradiance difference betweendirect sunlight and indoor illumination. Conven-tional solar cells are also inflexible (rigid), whichmakes it difficult to attach them to non-rigiditems such as clothing and paperback books.

An emerging and less explored option is solarenergy harvesting based on organic semiconduc-tors [3] (an array of solar cells we recentlydesigned is shown in Fig. 2a). With this technol-ogy, solar cells can be made flexible. Moreover,organic semiconductor-based panels operatewith constant efficiencies over different bright-ness levels. However, their efficiency is typically1–1.5 percent, which is much lower than the effi-ciency of conventional inorganic solar panels.One of the limitations of organic photovoltaicelements is that they are reactive with watervapor and oxygen. There are several approachesto overcoming this in a flexible, thin film archi-tecture, including encapsulation against environ-mental degradation. We have demonstrated a

flexible, thin film package capable of protectingthese solar cells in ambient environments [11].We have also recently developed a solar cellarchitecture which uses an organic semiconduc-tor (bis-thienyl pentacene) that has a metastableoxidized state. Even if the device package is pen-etrated and the material oxidizes, the materialreturns to its reduced form upon illuminationand the cell recovers its performance [12].

To put the energy availability numbers in per-spective, consider a system with a 10cm2 organicsemiconductor cell. Outdoors, the system willharvest 10cm2 ⋅ 100mW/cm2 ⋅ 0.01 = 10mW.Under the assumption that receiving a bit2

requires 1nJ, the achievable data rate will be (10⋅ 10–3)/(1 ⋅ 10–9) =10 Mb/s. The achievable datarate with indoor lighting will be (10–5)/(1 ⋅ 10–9)= 10 kb/s.

Piezoelectric Harvesting — Another potential sourceof energy is piezoelectric (motion) energy. It canbe generated by straining a material (e.g.,squeezing or bending flexible items). An exam-ple is energy harvesting through footfall, where aharvesting device is placed in a shoe and piezo-electric energy is generated and captured witheach step [13].

Piezoelectric harvesting is characterized bythe energy captured per actuation at a particularstrain (usually 1–3 percent). In [13] it was shownhow to harvest 4μJ/cm2 per deflection with astrain of approximately 1.5 percent from strainingpolyvinylidene fluoride (PVDF), a highly compli-ant piezoelectric polymer. Assume that a 10cm2

of material is employed in an environment whereit is strained once per second (that is, strain fre-quency of 1Hz). This would provide 10 ⋅ 4 ⋅ 10–6

= 40μW. If, similar to above, we assume thatreceiving a bit costs 1nJ, the bit rate that can besupported is (4.0 ⋅ 10–5)/(1 ⋅ 10–9) = 40 kb/s.

ENERGY STORAGEWithout the ability to store energy, a device canoperate only when directly powered by environ-mental energy. For a tag, energy storage compo-nents need to be compact and efficient, andneed to have very low self-discharge rates.

Figure 2. a) An organic semiconductor-based small-molecule solar cell [12]; (b) a thin-film printable battery, both developed in theColumbia Laboratory for Unconventional Electronics (CLUE).

(a) (b)

2 Reception is moreexpensive than transmis-sion.

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 20

IEEE Wireless Communications • December 2010 21

Rechargeable batteries are an excellent optionfor energy storage, and numerous battery optionsare available. Thin film batteries are particularlyattractive for EnHANTs since they can be madeflexible (a prototype we recently designed isshown in Fig. 2b). Note that a battery needs tobe supplied with a voltage exceeding the internalchemical potential in order to start storing pro-vided energy. This implies that charge generatedat a low voltage, such as that possibly producedduring low harvester excitation (for example,when a solar cell is located in a dimly lit place),cannot be stored without voltage upconversion.However, organic solar cells discussed above canproduce relatively high voltage even at low exci-tation. This highlights another advantage ofusing organic solar cells for EnHANTs: in a sys-tem using an organic solar cell, a battery canstore energy when illumination is low, while in aconventional system a battery would not storecharge produced at that illumination level.

Capacitors can also be used for energy stor-age. Capacitors can receive any charge whichexceeds their stored voltage and be cycled manymore times than batteries. The disadvantage ofusing capacitors, however, is that as a capacitorgets more charged, it becomes more difficult toadd charge, and large electrolytic capacitors self-discharge over hours or days. The energy density(how much energy can be stored per unit of vol-ume) of capacitors is also much lower. A typicalbattery can store about 1000J/cm3, whereas highperformance ceramic capacitors can store 1-10J/cm3.

LOW POWER COMMUNICATIONS

Ultra-wide band (UWB) impulse radio (IR) is acompelling technology for short range ultra-low-power wireless communications. It uses veryshort pulses (on the order of nano-seconds) thatare transmitted at regular time intervals with thedata encoded in the pulse amplitude, phase, fre-quency, or position. At low data rates, the shortduration of the pulses allows most circuitry inthe transmitter or receiver to be shut downbetween pulses, resulting in significant powersavings compared to narrow-band systems.

Practical CMOS IR circuits with energy con-sumption on the order of a nano Joule per bithave recently been demonstrated. For example,in [4] we demonstrate a UWB receiver and trans-mitter that require 1.65nJ/bit and 280pJ/bit,respectively, at a data rate of 1 Mb/s.3 Our ongo-ing research indicates that UWB-IR transceiversin the 3-5GHz band with data rates of 0.1–1Mb/s and receiver and transmitter consumptionof less than 500pJ/bit and 50pJ/bit, respectively,are within reach.

In this section, we outline the envisioneddesign of UWB transceivers for EnHANTs,highlight UWB features that are important forhigher-layer protocols, and briefly discuss higherlayer protocol challenges.

ENERGY COSTS — A PARADIGM SHIFTAn important paradigm shift associated withultra-low-power IR-UWB transceivers is that theenergy to receive a bit is much higher than theenergy to transmit a bit. This is significantly dif-

ferent from traditional technologies. In large-scale systems the energy to transmit is higherthan the energy to receive, due to transmitter’sneed to generate a signal with enough power toreach long distances. In GSM cellular phones,for example, the transmitter power is on theorder of Watts, and the receiver power is on theorder of hundreds of mW. In shorter-range sys-tems the energy to transmit and to receive areon the same order of magnitude. For example,for Bluetooth and IEEE 802.15.4, the transmitand the receive power are on the order of tensof mW.

In IR-UWB, transmitter energy consumptionis further reduced. In conventional systems, nar-row-band modulated sinusoids are transmitted,and the transmitter has to be active for theentire duration of the signal transmission. Asmentioned above, in UWB very short pulsesconvey information, so the transmitter andreceiver can wake up for very short time inter-vals to generate and receive pulses, and cansleep between subsequent pulses. The receiver’senergy is mostly spent on running low-noiseamplification and data-detection circuits that areconsuming energy whenever the device listens tothe medium. Hence, there is no difference (interms of energy) between receiving informationand listening to the medium.

The new energy trade-offs call for the designof novel protocols for EnHANTs, since manylegacy algorithms were developed under theassumption of transmission being more expen-sive than reception.

INACCURATE CLOCKSAccurate on-chip clocks cannot be powereddown and thus consume a lot of energy. Hence,they have to be avoided to achieve ultra-low-power operation. One viable solution is to useenergetically cheap clocks (e.g., clocks availablefrom ultra-low-power ring oscillators). However,the frequency of such clocks will vary significant-ly from tag to tag and its stability over time isalso poor.

A UWB receiver has to wake up at certaintimes in order to receive pulses. Determiningthese times with inaccurate clocks imposes majorchallenges, so while inaccurate clocks save ener-gy, they increase the energy spent on reception.Moreover, traditional low-power sleep-wake pro-tocols heavily rely on the use of accurate timeslots. Hence, eliminating the availability of accu-rate clocks in a tag requires redesigning proto-cols that were originally designed specifically forenergy efficient networking.

DESIGN ALLOWING FOR A HIGH-POWER MODEEnHANTs transceivers employ numerous designfeatures targeted at decreasing energy consump-tion as much as possible. However, in somecases it may be beneficial to spend more energythan what is typically spent by a tag (i.e., whenthe battery is fully charged and the tag is har-vesting energy). In such cases an EnHANT canoperate in a high-power mode. To enable suchmodes, EnHANT circuitry can be designed tocontain optional power-hungry hardware mod-ules that would allow, for example, for a moreaccurate/faster clock to be turned on. Other

The new energytrade-offs call for the

design of novel protocols for

EnHANTs, sincemany legacy

algorithms weredeveloped under the

assumption of transmission being

more expensive than reception.

3 Energy consumption ismeasured at a particulardata rate, since per-bitparameters differ at differ-ent data rates. Note thatat lower bit rates, the ener-gy per bit can increase dueto the impact of fixed rate-independent circuitry.

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 21

IEEE Wireless Communications • December 201022

components that could be considered are moresensitive receiver stages and more selective fil-ters. Realizing the benefits of high-power modesis an interesting design challenge for higher-layer protocols.

ADDITIONAL CHALLENGES IN ENHANTCOMMUNICATIONS

Similar to traditional sensor networking, in pair-wise EnHANT communications three states canbe identified: independent (tags not communicat-ing), paired (synchronized and exchanging peri-odic keep-alive bursts to stay synchronized), andcommunicating (sending and receiving data). Tocontrol its energy spending, a tag can movebetween states with respect to each of its neigh-bors. In addition, in each state a tag can spendenergy at different rates.

Communicating EnHANTs need to coordi-nate their transmissions in order to ensurethat they do not run out of energy. Note thatsince listening is energetically expensive, thecommunication rate of a tag that transmitsinformation is not only related to tag’s ownenergy, but is also tightly coupled with theenergy of a tag that receives the information.To make joint decisions on communicationrates, EnHANTs need to exchange informa-t ion about their energy states . However,exchanges of complete energy parameters maybe too costly. Determining how much infor-mation EnHANTs should exchange and decid-ing on the frequency of information updatesremains an open problem.

Since it is energetically cheaper for a tag totransmit than to listen, if a tag is very low onenergy, it could transmit pulses but not listen tothe medium. Accommodating the presence ofsuch transmit-only tags is a topic for furtherresearch.

A benefit of the harvesting system is thatEnHANTs in close proximity will be subject tocommon stimuli through their energy harvestingchannels. Examples include lights turning on/offor running with modulated intensity (e.g., the60Hz variation in fluorescent lighting), or vibra-tions felt by more than one tag. For instance,

when a light is turned on, a tag can assume thatthe energy parameters of all its neighborschange, and behave accordingly. Informationabout the relative similarities or differencesbetween EnHANTs energy stimuli can be usedfor synchronization via a channel which is effec-tively always open.

Finally, security and privacy are very impor-tant issues for the proposed EnHANTs applica-tions. Lightweight techniques that have beendesigned for sensor networks and for RFIDs canserve as starting points in EnHANTs securityresearch.

ENHANT ENERGY SPENDING AS ALARGE-SCALE OPTIMIZATION PROBLEM

When communicating with each of its neighbors,a tag decides on the state of communication andthe rate of energy consumption. When manytags are involved, the joint decisions on statesand rates form a large-scale optimization problem,for which a suitable solution needs to be calcu-lated by low-power EnHANTs without extensiveexchange of control information.

The complexity of tag’s energy harvestingsystem needs to be captured in a relatively sim-ple way. A possible abstraction is to character-ize the energy harvesting system by themaximum energy storage capacity C (Joules),the currently available energy level E (Joules),the energy charge rate r (Watts), and the ener-gy consumption rate e (Watts). Note that theenergy charge rate r depends both on the har-vesting rate and the properties of the energy stor-age. For example, when a battery is used, r ispositive only when the voltage at the energyharvesting component exceeds the internalchemical potential of the battery. When acapacitor is used, the relationship of r andenergy harvesting rate varies with E. The r val-ues of different EnHANTs operating in thesame environment will be significantly different.Our experiments with commercial hardwareshow that harvesting rates of identical harvest-ing devices under identical light conditions candiffer by about 30 percent.

Figure 3. Indoor energy availability measurements: a) total energy available per day in two different locations over 15 days and b) irradi-ance in two different locations over 7 days.

Day

(a)

50

1

0

Ener

gy a

vaila

ble

(J/c

m2 /

day)

2

3

4

5

10 15Day

(b)

10

Irra

dian

ce (μ

W/c

m2 ) 300

400

200

100

2 3 4 5 6 7

Office AOffice B

Office COffice D

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 22

IEEE Wireless Communications • December 2010 23

Inventory control theory [14] can potentiallybe used to formulate and solve the optimiza-tion problem. An energy-harvesting tag canbe viewed as a manufacturing system com-posed of a factory and a warehouse. Considerthe above-identified system parameters C, r,e , and E . A factory-warehouse has a finitecapacity, a rate at which products are manu-factured, a rate at which products are pur-chased, and a level of inventory — al l ofwhich directly correspond to the harvestingsystem parameters. A warehouse inventorymanagement system strives to ensure that thedemand is met and that there are no short-ages. Similarly, the tag’s energy managementsystem should ensure that it utilizes the avail-

able resources efficiently and does not runout of energy.

The similarities with inventory systems allowfor simple modeling of a stand-alone EnHANT.A network of EnHANTs calls for extensions ofexisting inventory management models. A par-ticular challenge in EnHANTs is the tight cou-pling of energy spending of different tags. AnEnHANT network can be viewed as a networkof factories in which the behavior of one signifi-cantly affects the behavior of the others. Fur-ther, while in a manufacturing system the centralcontroller has complete knowledge, in a net-work of tags, distributed low complexity algo-rithms using partial knowledge will have to beemployed.

Figure 4. Irradiance measurements in different parts of a bookshelf: a) the floor plan of the office where themeasurements were taken; b) the leftmost stall (stall A) in which measurement were taken with measure-ment locations; and c) the irradiance values in different locations on the bookshelf (measurements takenwhen outdoor light does not affect the values).

Boookshelf indexBA

10

0

Irra

dian

ce (μ

W/c

m2 )

20

30

40

50

C D

(c)

E

Shelf 7 -- topShelf 6Shelf 5Shelf 4Shelf 3Shelf 2Shelf 1 -- bottom

(a) (b)

Bookshelf

A B C D E

Overheadlights

When communicat-ing with each of its

neighbors, a tagdecides on the state

of communicationand the rate of

energy consumption.When many tags are

involved, the jointdecisions on states

and rates form a large-scale opti-mization problem.

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 23

IEEE Wireless Communications • December 201024

ENERGY HARVESTING EXPERIMENTS &ENHANT TESTBED DEVELOPMENT

It is important to quantify how much energy canbe harvested indoors, and the spatio-temporalvariability. To the best of our knowledge, noextensive long-term indoor light energy measure-ment studies have been conducted. In this sec-tion, we outline the development of anexperimental EnHANTs testbed, and briefly dis-cuss some of our energy harvesting measure-ments.

ENHANT TESTBED DEVELOPMENTExisting wireless mote designs, mostly optimizedfor wireless sensor applications, do not permitexperimentation with physical layer communica-tions protocols. Hence, we are in the process ofbuilding EnHANT prototypes.

Our current first-phase prototypes mostly usecommercial off-the-shelf (COTS) communica-tion and energy harvesting hardware compo-nents. One current prototype is based on aMICA2 mote attached to a small battery and acustom flexible solar cell [12] that charges thebattery, and to a TAOS TSL230rd light-to-fre-quency converter for light measurements. Theseprototypes serve as a platform for further cus-tom hardware integration and for experimentson energy harvesting aware communication pro-tocols [15]. We are currently replacing the COTScomponents with custom-designed hardware,namely, a thin film printable battery similar tothe one shown in Fig. 2, and a custom-designedUWB transceiver [4].

INDOOR LIGHT ENERGY MEASUREMENTSSince June 2009 we have been conducting a lightmeasurement study in office buildings in NewYork City. We continuously record irradiance inseveral different locations using TAOS TSL230rdlight-to-frequency converters installed on Lab-Jack U3 Data Acquisition boards. For each loca-tion, we calculate from these measurements thetotal energy available to a 1cm2 solar cell in aday (denoted by Ed).4 For example, Fig. 3a showsthe values of Ed in two different locations over15 days (Jan. 21, 2010–Feb. 6, 2010). There aresignificant differences between the energy avail-abilities at the different locations: on average, theLabJack boards receive 2.9J/cm2/day and1.1J/cm2/day in Office A and B, respectively. Thedifference stems from different office locationsand layouts as well as occupants’ behaviors (e.g.,raising the blinds). Assuming a 10cm2 solar cellwith a 1 percent overall conversion efficiencyand a bit communication cost of 1nJ/bit, theaverage achievable data rate is Ed ⋅ 10cm2 ⋅0.01/(24 ⋅ 3600s/day ⋅ 1 ⋅ 10–9J/bit). Accordingly,the average achievable continuous bit rates are3.35 kb/s for a tag in Office A and 1.27 kb/s for atag in Office B.

As can be seen in Fig. 3a, for a particularlocation, the available energy also varies greatlybetween different days (the standard deviationsare 0.8 and 0.56 in Office A and B, respectively),due to factors such as weather conditions, indoorlights use, and shading use. In addition to thetemporal variations between days, there may be

significant temporal variations within a day. Thecharacteristics of these variations depend on thelocation. For instance, Fig. 3b shows irradiancelevels recorded over 7 days (Jan. 31, 2010–Feb.6, 2010) in two different offices. The shapes ofthe two curves are quite different. In Office C,the LabJack board gets most of the energy fromsunlight, resulting in variations related to theSun’s diurnal cycle. In Office D, the LabJackboard gets most of the energy from artificiallight sources and is exposed to direct sunlight foronly a short duration. This results in a curve thatis flat during most of the work day and has minortemporal variations.

The measurements also indicate that evenwhen there is little temporal variability (e.g.,within an office il luminated by artificialsources), there is still high spatial variability.For example, in Fig. 4 we demonstrate how theirradiance levels vary as a function of the dis-tance (and orientation) from light sources. Wemeasured the irradiance levels at different loca-tions on a bookshelf which consists of 4 stallswith 7 shelves in each, and is illuminated byoverhead fluorescent lights as shown in Fig. 4a.We took measurements at the intersectionpoints of shelves and bookshelf stalls, as depict-ed in Fig. 4b. Figure 4c shows the measuredirradiance levels for each of the shelves. Itdemonstrates that the available energy levelsare significantly different even within a seem-ingly uniform environment.

The communications and networking proto-cols need to be designed such that they can copewith this high spatial variability as well as withthe spatial and temporal variations demonstratedin Fig. 3.

CONCLUSIONS

Ultra-low-power Energy Harvesting Active Net-worked Tags (EnHANTs) are enablers for a newtype of a ubiquitous wireless network that lies inthe domain between sensor networks andRFIDs. While RFIDs make it possible to identi-fy an object in proximity to a reader, EnHANTsmake it possible to search for an object in a net-work of devices and continuously monitorobjects’ locations and proximities to each other.EnHANTs will enable the Internet of Things byproviding infrastructure for novel tracking appli-cations.

EnHANTs necessitate rethinking of commu-nications and networking principles, and requirecareful examination of the particularities ofultra-low-power and energy harvesting technolo-gies. We outlined several important characteris-tics of EnHANTs and described harvesting andcommunications devices composing them. Wehave also described some of our indoor lightmeasurements. We have shown that the natureof EnHANTs requires a cross-layer approach toenable effective networking between deviceswith severe power and harvesting constraints.

ACKNOWLEDGMENTSThis work was supported in part by the Voda-fone Americas Foundation Wireless InnovationProject, Google Inc., NSF grants CNS-0916263and CCF-0964497, and by an NSERC CGS

Existing wirelessmote designs, mostly optimized forwireless sensor applications, do notpermit experimenta-tion with physicallayer communica-tions protocols.Hence, we are in theprocess of buildingEnHANT prototypes.

4 Ed is the energy avail-able for harvesting. Aspreviously noted, anorganic semiconductor-based solar cell would beable to collect 1–1.5 per-cent of the available ener-gy.

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 24

IEEE Wireless Communications • December 2010 25

grant. We thank Enlin Xu, Michael Zapas, andMatthias Bahlke for their assistance with thelight measurements.

REFERENCES[1] M. Gorlatova et al., “Challenge: ultra-Low-Power Ener-

gy-Harvesting Active Networked Tags (EnHANTs),” Proc.ACM MobiCom’09, Sept. 2009.

[2] J. Paradiso and T. Starner, “Energy Scavenging forMobile and Wireless Electronics,” IEEE Pervasive Comp.,vol. 4, no. 1, pp. 18–27, 2005.

[3] P. Peumans, A. Yakimov, and S. Forrest, “Small Molecu-lar Weight Organic Thin-Film Photodetectors and SolarCells,” J. Appl. Phys., vol. 93, 2003, p. 3693.

[4] M. Crepaldi et al., “An ultra-Low-Power Interference-Robust IR-UWB Transceiver Chipset Using Self-Synchro-nizing OOK Modulation,” Proc. IEEE ISSCC’10, Feb.2010.

[5] I. Ehrenberg, C. Floerkemeier, and S. Sarma, “InventoryManagement with an RFID-Equipped Mobile Robot,”Proc. IEEE CASE’07, Sept. 2007.

[6] A. Kansal et al., “Power Management in Energy Har-vesting Sensor Networks,” ACM Trans. EmbeddedComp. Sys., vol. 6, no. 4, 2007, p. 32.

[7] K. Kar, A. Krishnamurthy, and N. Jaggi, “Dynamic NodeActivation in Networks of Rechargeable Sensors,”IEEE/ACM Trans. Net., vol. 14, no. 1, 2006, pp. 15–26.

[8] K.-W. Fan, Z. Zheng, and P. Sinha, “Steady and FairRate Allocation for Rechargeable Sensors in PerpetualSensor Networks,” Proc. ACM SenSys’08, Nov. 2008.

[9] J. Taneja, J. Jeong, and D. Culler, “Design, Modeling,and Capacity Planning for Micro-Solar Power SensorNetworks,” Proc. IEEE IPSN’08, Apr. 2008.

[10] J. Randall, Designing Indoor Solar Products, 1st Ed.,Wiley, 2005.

[11] M. E. Bahlke, S. P. Subbarao, and I. Kymissis, “A Labo-ratory Process for Encapsulation of Air Sensitive Organ-ic Devices,” IEEE Trans. Electron Devices, vol. 57, no. 1,Jan. 2010, pp. 153–56.

[12] A. A. Gorodetsky et al., “Solar Cell from a SolutionProcessable Pentacene with Improved Air Stability,”Chem. Mater., vol. 21, no. 18, 2009, pp. 4090–92.

[13] J. Kymissis et al., “Parasitic Power Harvesting inShoes,” Proc. 2nd Int’l. Symp. Wearable Computers,1998.

[14] D. Simchi-Levi, X. Chen, and J. Bramel, The Logic ofLogistics, 2nd Ed., Springer, 2005.

[15] M. Gorlatova et al., “Demo: Prototyping Energy Har-vesting Active Networked Tags (EnHANTs) with MICA2motes,” Proc. IEEE SECON’10, June 2010.

BIOGRAPHIESMARIA GORLATOVA (mag2206@columbia.edu) is a Ph.D. Can-didate in the Columbia University Electrical EngineeringDepartment. Her research interests are in networking ultra-low-power wireless devices. She has received her B.S. andM.S. degrees from the University of Ottawa, Canada. Priorto joining Columbia’s Ph.D. program, Maria worked as aresearch scientist at Defence R&D Canada and at TelcordiaTechnologies. Maria is a recipient of Columbia UniversityPresidential Fellowship and Canadian Graduate Scholar(CGS) NSERC Fellowships.

PETER R. KINGET (kinget@ee.columbia.edu) received an engi-neering degree in Electrical and Mechanical Engineering

and the Ph.D. in electrical engineering from the KatholiekeUniversiteit Leuven, Belgium. He has worked in industrialresearch and development at Bell Laboratories, Broadcom,Celight and Multilink before joining the faculty of theDepartment of Electrical Engineering, Columbia University,in 2002. His research interests are in analog and RF inte-grated circuits and signal processing using nanoscaleCMOS technologies. He has been an Associate Editor ofthe IEEE Journal of Solid State Circuits (2003–2007) andthe IEEE Transactions on Circuits and Systems II(2008–2009). He is also serving on the Technical ProgramCommittees of the International Solid-State Circuits Con-ference and the European Solid-State Circuits Conference.He is a “Distinguished Lecturer” for the IEEE Solid-StateCircuits Society.

IOANNIS (JOHN) KYMISSIS (johnkym@ee.columbia.edu) is anAssistant Professor of Electrical Engineering at ColumbiaUniversity. His research focuses on the processing, design,and applications of thin film semiconductors to sensorsand actuators, with a particular emphasis on the use oforganic semiconductors and recrystallized silicon. Johngraduated with his SB, M.Eng., and Ph.D. degrees fromMIT, and is a senior member of the IEEE.

DAN RUBENSTEIN (danr@cs.columbia.edu) is an AssociateProfessor in the Department of Computer Science atColumbia University. He received a Ph.D. in Computer Sci-ence from University of Massachusetts, Amherst. Hisresearch interests are in network technologies, applica-tions, and performance analysis. He is an editor forIEEE/ACM Transactions on Networking, program chair ofACM Sigmetrics 2011, and has received an NSF CAREERAward, IBM Faculty Award, the three paper awards fromACM SIGMETRICS 2000, IEEE ICNP 2003, and ACM CoNext2008.

XIAODONG WANG [S’98, M’98, SM’04, F’08](wangx@ee.columbia.edu) is a Professor of Electrical Engi-neering at Columbia University. He received the Ph.D.degree in Electrical Engineering from Princeton University.His research interests fall in the general areas of comput-ing, signal processing and communications, and he haspublished extensively in these areas. Among his publica-tions is a recent book entitled “Wireless CommunicationSystems: Advanced Techniques for Signal Reception”, pub-lished by Prentice Hall in 2003. He received the 1999 NSFCAREER Award, and the 2001 IEEE Communications Soci-ety and Information Theory Society Joint Paper Award. Hehas served as an Associate Editor for the IEEE Transactionson Communications, the IEEE Transactions on WirelessCommunications, the IEEE Transactions on Signal Process-ing, and the IEEE Transactions on Information Theory. Heis a Fellow of the IEEE and listed as an ISI Highly-citedAuthor.

GIL ZUSSMAN received his Ph.D. degree in electrical engi-neering from the Technion in 2004. Between 2004 and2007 he was a postdoctoral associate at MIT. He is anassistant professor of electrical engineering at ColumbiaUniversity, and his research interests are in the area ofwireless networks. He is a recipient of the Fulbright andthe Marie Curie Outgoing International Fellowships, theIFIP Networking 2002 Best Student Paper Award, the ACMSIGMETRICS ’06 Best Paper Award, and the DTRA YoungInvestigator Award.

While RFIDs make itpossible to identify

an object in proximity to a

reader, EnHANTsmake it possible tosearch for an object

in a network ofdevices and

continuously monitorobjects’ locations

and proximities toeach other.

GORLATOVA LAYOUT 12/9/10 10:59 AM Page 25