an 86% efficiency 12 µw self-sustaining pv energy harvesting

1424 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 50, NO. 6, JUNE 2015

An 86% Efficiency 12 µW Self-Sustaining PV EnergyHarvesting System With Hysteresis Regulation and

Time-Domain MPPT for IOT Smart NodesXiaosen Liu, Student Member, IEEE, and Edgar Sánchez-Sinencio, Life Fellow, IEEE

Abstract—This paper presents a fully-integrated µW-levelphotovoltaic (PV) self-sustaining energy harvesting system pro-posed for smart nodes of Internet of Things (IOT) networks. Ahysteresis regulation is designed to provide a constant 3.3 V outputvoltage for a host of applications, including powering sensors,signal processors, and wireless transmitters. Due to the stringentpower budget in IOT scenarios, the power consumption of theharvesting system is optimized by multiple system and circuitlevel techniques. Firstly, the hill-climbing MPPT mechanismreuses and processes the information of the hysteresis controllerin the time-domain and is free of power hungry analog circuits.Secondly, the typical power-performance tradeoff of the hys-teresis controller is solved by a self-triggered one-shot mechanism.Thus, the output regulation achieves high-performance and yetlow-power operations. Thirdly, to execute the impedance tuningof MPPT, the capacitor value modulation (CVM) scheme is pro-posed instead of the conventional frequency modulation scheme,avoiding quiescent power consumption. Utilizing a commercialPV cell of 2.5 cm2, the proposed system provides 0–21 µW outputpower to the IOT smart nodes. Measured results showed that thePV harvesting system achieved both ultra-low power operationcapability at 12 µW and a peak self-sustaining efficiency of 86%.Index Terms—Capacitor value modulation, energy harvesting,

Internet of things, MPPT, photovoltaic, power management.

I. INTRODUCTION

W ITH recent developments in the microelectromechan-ical systems (MEMS) sensors and down-scaling of sil-

icon fabrication technology, the concept of Internet of Things(IOT) has been proposed to uniquely identify objects and theirvirtual representations in an Internet-like structure [1], [2] asillustrated in Fig. 1(a). In such a network, individual nodes,also called smart nodes, are often implemented as system-on-chip (SOC) solutions, containing sensors, signal processors andwireless transceivers. To power the nodes, multiple possible en-ergy sources are available such as photovoltaic [3], piezoelec-tric [4], thermoelectric [5], and RF [6]. Compared to these othercandidates, photovoltaic (PV) cells potentially provide a higherpower density and relatively smaller size. The output energy ofPV cells is commonly managed by DC-DC converters with off-chip inductors or transformers, featuring high power throughput

Manuscript received August 03, 2014; revised December 16, 2014; acceptedMarch 22, 2015. Date of publication April 21, 2015; date of current versionMay22, 2015. This paper was approved by Associate Editor Woogeun Rhee.The authors are with Texas A&M University, College Station, TX 77843

USA.Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/JSSC.2015.2418712

and efficiency [7]. However, full integration is preferable to theapplication of smart nodes, and high quality on-chip inductorsare not widely available for the CMOS technology. Alterna-tively, the monolithic switched capacitor topology is chosen toeliminate the need for an off-chip inductor [8].Creating an optimal output regulation for a harvesting system

is a difficult design challenge. The conventional DC-DC powermanagement theory, which assumes the energy source to beideal with constant voltage and possessing infinitely availablepower, is not valid in the PV scenario [9]. In fact, the PV energysource normally has a weak power density and cannot sustain astable output voltage under heavy loading conditions. To avoida loading effect on the harvesting system, [10] proposed a gatedoutput control based on a digital-to-analog converter (DAC) andcomparators, which consumed quiescent current and greatly re-duced power conversion efficiency. In this paper, an architecturedesign is proposed to solve power conversion problems. As il-lustrated in Fig. 1(b), this solution involves a hysteresis regula-tion to gate the conduction between a buffer capacitor and

, providing a constant output voltage. Here represents asupercapacitor or a battery. Thus, the hysteresis regulation guar-antees adaptive maximum power point (MPP) harvesting overvarious light intensities. When the available PV power is lowand not able to sustain its loads, the switch will be turnedoff and prevent the loads from draining off the charge pump.Thus, the harvesting system is always operated under MPP con-dition regardless of the illumination intensities. It can power ahost of applications in the smart nodes, including sensors, wire-less transmitters and battery chargers [11].Another practical challenge in a power harvesting design

is the fact that a PV energy source can experience changesin its power density; thus, its MPP depends on differentenvironmental variables such as illumination intensity andtemperature [12]. Therefore, a maximum power point tracking(MPPT) technique is needed to dynamically match the outputimpedance and constantly achieve maximum power transferunder those environmental variations. An MPPT circuit canbe one of the most power hungry blocks in the harvestingsystem. MPPT circuits require complicated signal processingcomponents such as a successive approximation register (SAR)or a digital signal processor (DSP) and can consume morethan 100 W in power [13], [14]. The hill-climbing MPPTalgorithm features the simplest mechanism and minimumdevices [15], which is favorable for monolithic and low powerpurposes. [16] developed a practical sample-and-hold (S/H)

0018-9200 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

LIU AND SÁNCHEZ-SINENCIO: AN 86% EFFICIENCY 12 W SELF-SUSTAINING PV ENERGY HARVESTING SYSTEM 1425

Fig. 1. (a) IOT smart nodes including the energy harvesting system and loads, (b) principle of the proposed power management.

Fig. 2. Proposed architecture of the energy harvesting system.

structure for the hill-climbing MPPT; however, it required apower-hungry analog current sensor and, thus, was not suitablefor microwatt-level energy harvesting. To avoid such issues,[10] monitored the output power with a DAC; however, it alsoincreased the circuit complexity, which consumed more power.In this paper, a time-domain hill-climbing MPPT is proposedto reuse the power information from former output regulation.Such a scheme eliminates the need for a current sensor or otheranalog circuits, and significantly reduces power consumption.Furthermore, the selection of an impedance tuning variable

for MPPT is also important for saving power. Theoretically,the input impedance of the charge pump relies on its switchingfrequency and capacitor value. Conventional approaches usea voltage-controlled oscillator (VCO) to continuously tune theswitching frequency [16], [10]. However, such a frequencymodulation scheme usually needs analog operational amplifiers(op-amps) with a quiescent power consumption that far exceedsthe stringent power budget for these applications. On the otherhand, a capacitor value modulation (CVM) does not requireanalog modules and can be implemented in the digital domain.Its drawback of consuming more chip area is relieved if theharvesting power is as low as tens of microwatts. For thisIOT application, low power is more critical than large on-chipcapacitors. CVM has been reported in [17]–[19] for dynamicoutput power scaling. In this paper, we propose a CVM ap-

proach for impedance tuning in MPPT, which consumes noquiescent power.This paper is organized as follows. Section II describes

the architecture of the proposed energy harvesting system.Section III details its implementation. The measured experi-mental results are shown in Section IV, and Section V concludesthe paper.

II. ADAPTIVE PV ENERGY HARVESTING SYSTEM

A. Architecture of the Proposed Energy Harvesting System

The architecture of the proposed adaptive energy harvestingsystem is shown in Fig. 2. The system consists of one forwardpath for energy delivery and two control paths: one for MPPTand one for regulation. In the forward path, as a direct interfaceto the PV cell, the charge pump boosts the PV output voltageto the required level and delivers the harvested PV powerto the loads. To achieve maximum power transfer, a dynamicMPPT algorithm is necessary to track the variable output resis-tance of the PV cell. The hill-climbing algorithm of MPPT,which is executed by the MPPT loop, as shown in Fig. 2, ischosen to minimize the hardware complexity and power con-sumption [15]. Once the sensed power information is available,a finite-state machine (FSM) executes the hill-climbing algo-rithm to tune the equivalent resistance of the charge pump


Fig. 3. (a) Generic structure of the the nested voltage tripler built with two voltage doublers, (b) macromodel of the charge pump, and (c) the programmablecapacitor bank.

and searches for the MPP. As will be discussed in Section II-B,is defined by the inverse product of switching frequency

and capacitor values . To tune , an energy-efficient CVMscheme is proposed instead of the conventional frequency mod-ulation scheme, avoiding quiescent power consumption.When illumination is weak, the PV source cannot contin-

uously power the loads of the smart nodes. To avoid such aharmful loading effect, the loads are periodically turned on andoff; thus, they cannot provide as a continuously availableparameter for sensing and regulation. As a solution, we proposeto sense the pumped voltage of the charge pump as illus-trated in Fig. 2; hence, a feedforward path of hysteresis regula-tion is applied to directly manage to power the loads. Theharvested PV energy is temporarily stored in a buffer capacitor

at the output of the charge pump. By comparing its voltagewith a high reference and a low reference , the reg-

ulation scheme turns on and off the switch, respectively,as shown in the inset of Fig. 2. The values of and are de-termined by the requirement of the loads. As a result, this feed-forward path does not rely on , preventing the loads fromdraining off the harvesting system.As will be discussed later, the one-shot hysteresis controller

in Fig. 2 utilizes less complex circuits than conventional PWMschemes. Its quiescent power consumption can be eliminatedand fitted for ultra-low power application. The controller canalso be reused for time-domain MPPT, where the rising time

is inverse of the harvested PV power . Thus, the powerinformation of the PV cell can be quantified and efficiently pro-cessed in the time-domain through a time-to-digital converter

(TDC), avoiding the power hungry current sensor. The sensingtheory is detailed in Section II-D.

B. 3 Charge Pump With CVMThe 3.3 V LVTTL is a widely used standard in commercial

ICs for the smart nodes. A one-stage doubler only boosts a PVvoltage of 1.1–1.5 V to a maximum of 3 V, which is notenough to drive loads with standard LVTTL of 3.3 V. In prin-ciple, the voltage doubler is a circuit that adds two input volt-ages. We can use this fact to implement by nestingtwo voltage doublers as shown in Fig. 3(a). The second dou-bler has one modified input from the PV operating voltage, ,resulting in a total boosting gain of 3. Note that the structurecan also be viewed as two 3 Dickson charge pumps driven bycomplementary clocks. Its operating waveforms are shown inFig. 3(a),1) When (logic) and (logic): ,

, and are turned off. , , are turned on., , and is charged to through

.2) When and : , , and

are turned off. , , are turned on. ,, and is charged to through .

The charge pump can be modeled as an ideal 1:3 DC trans-former in series with an equivalent resistor [19] as shownin Fig. 3(b). can be calculated as

(1)


Fig. 4. (a) Architecuture of the hysteresis controller and (b) the operating waveforms.

where is the charge multiplier coefficient [23] and equals0.5. According to (1), is determined by two parameters:the switching frequency and the switching power capacitor

. The generation of variable usually needs complex aux-iliary circuits and consumes quiescent power, thereby affectingthe efficiency of the harvester system [16]. Therefore, the CVMapproach of tuning is chosen to avoid quiescent power asshown in Fig. 3(c), where consists of a fixed part andparallel capacitors of value . By switching the programmablecapacitor bank, the value changes from up to

.

C. Hysteresis Output Regulation

The block-level structure of the hysteresis controller inFig. 2 is depicted in Fig. 4(a). Its operating waveforms areshown in Fig. 4(b). is connected to through a passingswitch . is the command signal given by the hys-teresis regulation. Once is higher than , is turnedon to discharge the buffer capacitor toward the loads.Once is below , is turned off and is chargedby the 3 charge pump.Moreover, the conventional hysteresis detection in

Fig. 4(b) has a tradeoff between speed and power consumption.The rising time is a slow moving signal, which can easilybe detected by such a low speed latched comparator. However,the falling time is a fast moving signal, which requiresa fast comparator. Thus, we do not use the conventionalcomparator depicted in Fig. 4(a). An architectural solution isproposed via a self-triggered one-shot mechanism to control

. detection is fulfilled by a switched comparator forbetter harvesting efficiency; detection is achieved by a highspeed comparator, which is gated by a one-shot mechanismto limit its energy cost. Its mechanism is illustrated in Fig. 5.Firstly, is turned off. The charge pump output voltage is

, and is compared with to detect in a switchedcomparator without quiescent current. Its strobe clock,

, is depicted in Fig. 4(b). Once , a high

Fig. 5. Self-triggered one-shot mechanism of the hysteresis controller.

speed comparator and are turned on for .Because is quickly discharged to through ,

is detected as a one-shot time for , and isimmediately turned off to save power. The hysteresis controllerturns off and resets for the next detection. Such aself-triggered one-shot mechanism forms a fast loop and avoidsan overkilling GHz trigger clock.

D. Time-Domain Quantization for MPPTThe signal from the regulation module in Fig. 4(b) can

be reused to indicate illumination intensity. Intuitively, highlight intensity provides higher PV power and quickly charges

. Its charging time is shorter than with low lightintensity. can be defined by . It can be counted andquantified by a TDC as shown in Fig. 4(a) with the strobe clock

.


Fig. 6. Pseudo-static model of a PV cell and charge pump power converter.

The quantitative relationship between the charging timeand can be modeled via a steady-state assumption [21] asshown in Fig. 6. The PV cell is characterized by the simplifiedsingle-diode model [22] as a light-controlled current sourcewith a parallel diode and a series resistor .

(2)

where is photocurrent, is diode saturation current,is thermal voltage, is a PV series resistor provided by themanufacturer, is diode current, is diode voltage, andis PV output current.An equivalent resistor models the buffer capacitor

and its gated switch with the charging current . According tothe steady-state assumption, the ripple of between and

is neglected, and equals to as a constant voltage.Moreover, can be averaged during the entire charging period

and expressed as

(3)

Applying Kirchoff's voltage law at the input node of the DCtransformer,

(4)

Solving (2) and (4) permits the equation to only have :

(5)

Due to the PV's nonlinearity, such type of equation does nothave a closed-form solution. A possible solution can be obtainedusing an iterative scheme [24]. Thus, we approach the solutionin a different way. In the time-domain, the harvested powercan be simply represented by

(6)

Note that is inverse-proportional to the rising time .Therefore, we use to indicate the trend of , and convertthe MPPT power sensing problem into the time-domain withoutconventional power sensors [16]. The reused variable can besimplified from (3) and (4) as

(7)

Fig. 7. (a) Electrical characterisitcs of a PV cell under different illuminationconditions and (b) flow chart of the adaptive MPPT for the PV harvestingsystem.

Note that is still a non-closed-form expression of . Bysetting different , is changed and results in different .

will be recorded by the TDC in FSM, and reused by thehill-climbing MPPT as follows.


Fig. 8. Detailed proposed architecture of the energy harvesting system.

E. Hill-Climbing AlgorithmThe operation of a PV cell is depicted in Fig. 7(a). The max-

imum power point is determined by various factors includingillumination conditions and fabrication technology. Generally,

increases with increasing light intensity. A FSM exe-cutes the hill-climbing tracking flow as shown in Fig. 7(b). nrepresents the number of parallel connected capacitors in a pro-grammable bank. Once the MPPT procedure is triggered, theFSM initializes the charge pump with a low boundary voltage

. All capacitors in the bank are connected as , andreaches its minimum value at . The

first sensing state records the power information in the formof . Then one programmable capacitor, , is tentativelydisconnected as , and is increased through (1) as

. The second sensing state recordsthe new with the tentative . Based on the characteristicsof the PV cell, a different will cause different PV operatingvoltages, through (4). It will also change the harvested power

, which is a function of , as shown in (6). By comparingof neighboring two steps in Fig. 7(a), the finite-state machine

(FSM) gets the trend that the tentative tuning is improvingor degrading . If keeps decreasing, MPP is not capturedand the FSM examines the next value of the capacitor bank as

. Once stops decreasing, MPP is achieved and theFSM stops the searching procedure. Finally, the charge pump islocked at the optimal operating state with minimum andmaximum .Different from the conventional perturb & observe (P&O)

approach, the developed hill-climbing algorithm is unilateral.Therefore, it does not have a common stability problem as os-cillating around MPP [20]. Although the unilateral monotonichill-climbing algorithm is not as accurate as the P&O approachand suffers a small power loss from the PV cell, it has less com-plexity and saves power consumption of control circuits.

III. SYSTEM BLOCKS HARDWARE IMPLEMENTATION

The proposed energy harvesting system depicted in Fig. 2 canbe implemented as shown in Fig. 8. Its building blocks will bediscussed next.

A. Compact Nested Voltage DoublerIn Fig. 3, the switch transistors and are cross-con-

nected and self-switched. However, such architecture limits theturn-on voltage of NMOS transistors to less than , whichranges from 1.1 to 1.5 V. The low turn-on voltage drasticallyincreases the conduction resistance and degrades boosting effi-ciency. Furthermore, the self-switching transistors suffer fromshoot-through current, which ruins the conversion efficiency.Other coupled parasitic capacitors also affect the self-switchingand require additional damping branches [25]. To eliminatethese problems, we propose to break the cross-connected gatesof and , and drive them separately with thehigher supply voltage non-overlapping clock as shownin Fig. 9(a). The four drivers used in Fig. 3(a) are implementedwith transistor and . For the second stage, arereplaced with PMOS switches to allow conducting voltageas high as . The operating waveforms of the chargepump are depicted in Fig. 9(b). When (logic) and

(logic), and are charged to and 2Vs,respectively. is discharged to the output as . When

and , and are charged toand 2Vs, respectively. is discharged to as . When

and , all of the switches are turned offto prevent the shoot-through current. are generated byfollowing auxiliary circuits.As introduced in Section II-B, the switched capacitors are

programmable and split into fixed part and N programmablecapacitors [17]–[19]. In this design, representand its value is 18.8 pF. The programmable capacitor bank isimplemented as Fig. 10 with and 18.8 pF. Theswitches are implemented by transmission gates. The 15 iden-tical rows are controlled by thermometer code from a4 bit FSM controller, which is introduced in Section III-D.

B. Start-up and Auxiliary Bias CircuitTo eliminate external biases and realize the self-sustaining

feature for an energy harvesting system, a start-up and auxiliarybias circuit is proposed to provide supply voltages and drivingsignals once the PV cell is connected to the harvester. In Fig. 11,


Fig. 9. (a) Modified architecture of the nested charge pump power converter and (b) its operation with the non-overlapping clocks in complementary phases.

Fig. 10. Schematic of the programmable capacitor bank.

a current-starved 250 kHz ring oscillator operates with appliedPV voltage . Its output clock, , drives a non-overlap-ping signal generator in the right side, which eliminates theshoot-through current and improves converter efficiency. Dif-ferent from the conventional non-overlapping clock generator[26], a delay line is placed in the feedback path. Therefore, theforward drivers are designed to maximize their fan-out capa-

bility and minimize their power consumption. The non-overlap-ping time is tuned by the delay line independently.However, these circuits are directly supplied by , which is

not capable of driving the charge pump. Thus, an auxiliarythree-stage Dickson charge pump is used to generate a highersupply voltage as . It is driven by the fromthe ring oscillator. supplies a level shifter and generates

with amplitude. The shifted helps inFig. 9(a) to have a gate drive voltage of during the turn-onperiod. Additionally, the auxiliary charge pump providesto all control circuits as a power supply and body bias.

C. Hysteresis ControllerThe one-shot hysteresis controller proposed in Fig. 5 is im-

plemented as shown in Fig. 12(a). The utilized comparators(low power) and (high speed) are shown in Fig. 13. For theLVTTL standard, and are set as 3.15 V and 3.3 V, re-spectively. In Fig. 12(b), when and is off, thecharge pump keeps charging and continues rising.A low power latched comparator, , is clocked by a sensingclock with a frequency that is twice that of the chargepump switching signal . When is charged up to ,the period of ends, , and is turned on. As


Fig. 11. Start-up cirucuits and auxiliary bias circuits for self-sustaining.

Fig. 12. (a) Architecture of the self-triggered one-shot hysteresis controller, (b) when and the controller detects , (c) when and thecontroller detects .

shown in Fig. 12(c), is quickly discharged to the output. Ahigh speed comparator, , regulates the discharging timeas one-shot. Once , the detecting circuit turns off

and is asynchronously reset for next charging period.

D. Implementation of Finite-State Machine andTime-to-Digital ConverterAs illustrated in Fig. 2, the MPPT function is realized in

a FSM. Its detailed state transfer chart and time diagram are

shown in Fig. 14. The structure of the FSM is demonstrated inFig. 15. It is clocked by from the hysteresis controller. Onthe top level, every tentative searching step is executed in 16system clocks of a 4 bit FSM. Initially, four clock periods of the

state are used to settle the harvesting transient of dueto the capacitor value modulation. At the end of state, thebinary code 0100 for the clock captures the quantized numberof rising time . Then the binary code 0101 for clock tenta-tively disconnects one row of the capacitor bank in Fig. 10 and


Fig. 13. Structures of (a) the low power latched comparator and (b) thehigh speed amplifier .

Fig. 14. State transfer chart of the finite-state machine and its time diagram.

increases . The state uses another four clock periods tosettle with the new , and uses the binary code 1010 for theclock to capture the new . Subsequently, and arecompared for MPPT decision in the binary code 1011 for theclock. If , the controller should keep searching theMPP; if , MPP is already achieved, and the con-troller locks in this state. The remaining four clocks are sparedfor I/O communication with smart nodes. The TDC counters arebuilt with asynchronous-reset ripple counters for minimum de-vice cost and dynamic power.

IV. EXPERIMENTAL RESULTSThe adaptive PV harvester system is designed and fabricated

in standard 0.18 m CMOS technology. The die photo of thefabricated chip is shown in Fig. 16. The entire energy harvestingsystem occupies a silicon area of 1.5 mm 1.5 mm. Dual layermetal-insulator-metal (MIM) on-chip capacitors are used forthe monolithic integration of the capacitor bank. The measure-ment setup is illustrated in Fig. 17. This smart node includes atemperature sensor, a microcontroller and a wireless transceiverCC2500. In general, the proposed energy harvesting system canoperate with a supercapacitor and/or a compact 3.3 V man-ganese silicon lithium battery, which are only used as storagecomponents. When there is not enough PV energy, the har-vesting system stops operating, and the battery or supercapacitorsolely powers the smart node.

The transient measurements were carried out to verify thebehavior of the output regulation and the relationship betweeninput light power and the time-domain variable . To emu-late indoor illumination, various light intensities from 150 to600 lux were applied. The light acceptor was a small com-mercially available PV cell featuring a compact 2.5 cm size.The load was characterized by a potentiometer from 200 k to10 M paralleled with a 33 mF supercapacitor. The transientPV cell voltage and buffering output voltage are shownin Fig. 18. The relationship between light intensities and ischaracterized as follows: With a weak light intensity of 150 lux,the system needs more time as there are 14 quantized steps fora full capacitor charge. With a strong light intensity of 600 lux,the system only needs five quantized steps for a full capacitorcharge.When the light illumination is changing, the MPPT of is

shown in Fig. 19. MPPT operation is indicated by : Whenis low, MPPT is turned on. Initially, the PV cell is given

150 lux and is 1.14 V. Then the illumination is increasedto 300 lux, and the MPPT is manually triggered with an ini-tial value of 1.13 V and quickly reaches the MPP. Theachieved MPP for 300 lux is 1.21 V.The detailed dynamic MPPT performance was tested as

shown in Fig. 20 with 450 lux illumination. The external signaltriggers the MPPT module and initializes the capacitor

bank. The system begins to execute the hill-climbing MPPTprocedure as shown in Fig. 7. After seven tentative capacitorchanges, the system detects that the charging time cannotbe shorter than six steps, which means the MPP is alreadycaptured. Thus, the capacitor bank is locked, the MPPT moduleis turned off, and the controller works in its minimum powerconsumption mode. The is maintained around 3.3 Vwith reduced ripples smaller than 50 mV due to the 33 mFsupercapacitor.The practical driving ability of the harvesting system is vali-

dated for an IOT smart node. For saving energy, the IOT smartnode is operated in a periodic sample-per-seconds scheme. Asshown in Fig. 21, the sensor and transceiver are turned on foronly 35 ms, which is set as a 0.1% duty ratio of the wholeperiod. The microcontroller reads the sensed environmentaltemperature around 27 C with 0.1 C sensitivity and transmitsit through CC2500 with a 2.4 GHz RF signal. A computer witha RF receiver captures the sensed data from the IOT smartnode. The remaining 35 seconds are scheduled as an idle modefor the smart node circuits; however, the harvesting systemkeeps trickle charging the Lithium battery. From the transientfigure, the harvesting system provides a 3.3 V supply witha 410 mV overshoot. Owing to the feedforward hysteresisregulation, the load overshoot cannot load the PV source andharvesting system.The accuracy of MPP tracking was characterized through

sweeping tests. The MPPT module was disabled, and the ca-pacitor bank was programmed by an external computer throughthe I/O communication ports. The harvested power versus theprogrammed number is depicted in Fig. 22 under differentlight intensities from 150 lux up to 600 lux. As a comparison,


Fig. 15. Simplified structure of the finite-state machine (FSM) with the TDC function.

Fig. 16. Die photograph of the fabricated chip.

the dynamically captured MPP values are also annotated onthe plot. For all four cases, the harvesting system successfullyconverges at the global optimal point. The MPPT tracking effi-ciency is 99%.Although the digital MPPT approach significantly reduces its

power consumption, the overall efficiency mainly depends onthe dynamic pattern of the MPPT procedure. If the FSM ini-tiates the MPPT module after a long time, such as every sev-eral seconds, the energy loss during the fast MPPT procedureis negligible. When the FSM triggers the MPPT module every

0.3 second, the output power begins to decrease. However, theMPPT procedure can be dynamically triggered by the micro-controller and sensors in the load. When the sensors detect thatthe environmental illumination is rapidly changing, the trig-gering frequency is increased. If the illumination is stable orslow changing, the microcontroller seldom triggers the MPPTprocedure to save power. The end-to-end peak efficiency withactive MPPT versus different MPP is demonstrated in Fig. 22,in which the harvester maintains efficiencies greater than 78%with output power above 8 W. Further increasing the illumi-nation intensity will induce higher , which deviates from

due to the use of the 3 charge pump and suffers thecharge redistribution loss [29].The detailed power consumption of the proposed system is

shown in Fig. 23. Due to the digital feature of the control cir-cuits, the entire power consumption with active MPPT is as lowas 294 nW.Table I compares the performance of the proposed work with

other state-of-the-art MPPT harvesters. This harvester useson-chip switched capacitors and features monolithic integra-tion. The peripheral circuits, including the hysteresis controller,FSM and the clock generator, are all powered by the harvesterand auxiliary charge pump. Thus, the entire harvesting systemis self-sustaining and needs no external bias. For controlledtrickle charging, the output voltage is regulated at 3.3 V with a150 mV ripple. The measured harvested power ranges from 0 to21 W depending on the illumination condition. For ordinaryoperation where the incoming dim indoor light is 420 lux andthe MPPT module is operated in the active mode, the dynamicoverall efficiency can achieve a peak value of 86.4% whiledelivering 12 W of throughput power. The proposed har-vester achieved a superior performance compared to reported


Fig. 17. Testing setup for trickle charging a IOT smart node.

Fig. 18. Different charging time under (a) 150, (b) 300, (c) 450 and (d) 600 lux conditions.

Fig. 19. Transient MPPT with illumination changing from 150 lux to 600 lux.

results, which could only achieve good efficiencies with alarge amount of PV power in the hundreds of microwatts, or

harvest a small amount of power below W but with poorefficiency [27], [3]–[8]. In summary, this PV energy harvestingsystem achieved both ultra-low power capability and excellentself-sustaining efficiency at the same time.

V. CONCLUSIONThis paper proposes a monolithic highly efficient W-level

photovoltaic energy harvesting system targeted for smart nodesin IOT networks capable of powering various loads such assensors, signal processors and wireless transceivers. Due to thestringent power budget in IOT scenarios, the power consump-tion of the harvesting system was optimized by the proposed ar-chitecture and circuit level innovations. First, the hill-climbingMPPT mechanism reused and processed the information of thehysteresis controller in the time-domain and was free of powerhungry analog circuits. Second, the power-performance tradeoff


TABLE IPERFORMANCE COMPARISON OF PV ENERGY HARVESTING SYSTEMS WITH MPPT

Fig. 20. Transient and waveforms during the MPPT procedure with450 lux illumination.

Fig. 21. Driving performance for an IOT smart node operation.

of the hysteresis controller was solved by the proposed self-trig-gered one-shot mechanism, allowing the output regulation toachieve high-performance and yet low-power operations. Third,the CVM scheme was proposed instead of the conventional fre-quencymodulation scheme, avoiding high quiescent power con-sumption. The harvesting system, fabricated in 0.18 m CMOS

Fig. 22. (Left) Output power with different capacitor values n under differentlight intensities and correspondingMPPs, and (right) end-to-end peak efficiencywith MPPT vs. different PV power.

Fig. 23. Detailed power consumption of the PV energy harvesting system.

technology, was tested on a temperature sensing smart node in-cluding a sensor, microcontroller, and a wireless transceiver.


The system provided a 3.3 V regulated voltage and achieved anend-to-end efficiency of 86.4% with a throughput power as lowas 12 W. Start-up and auxiliary bias circuits were also imple-mented to provide a self-sustaining operation when the systemwoke up from a completely dark environment.

REFERENCES

[1] R. Yao, W. Wang, M. Farrokh-Baroughi, H. Wang, and Y. Qian,“Quality-driven energy-neutralized power and relay selection forsmart grid wireless multimedia sensor based IoTs,” IEEE Sensors J.,vol. 13, pp. 3637–3644, Oct. 2013.

[2] F. Maciá-Pérez, F. Mora-Gimeno, D. Marcos-Jorquera, J. A.Gil-Martínez-Abarca, H. Ramos-Morillo, and I. Lorenzo-Fonseca,“Network intrusion detection system embedded on a smart sensor,”IEEE Trans. Ind. Electron., vol. 58, pp. 722–732, Mar. 2011.

[3] H. Kim, S. Kim, C. Kwon, Y.Min, C. Kim, and S. Kim, “An energy-ef-ficient fast maximum power point tracking circuit in an 800- W pho-tovoltaic energy harvester,” IEEE Trans. Power Electronics, vol. 28,pp. 2927–2935, Jun. 2013.

[4] P. Gasnier, J. Willemin, S. Boisseau, G. Despesse, C. Condemine, G.Gouvernet, and J.-J. Chaillout, “An autonomous piezoelectric energyharvesting IC based on a synchronous multi-shot technique,” IEEE J.Solid-State Circuits, vol. 49, pp. 1561–1570, Jul. 2014.

[5] S. Carreon-Bautista, A. Eladawy, A. N. Mohieldin, and E.Sanchez-Sinencio, “Boost converter with dynamic input impedancematching for energy harvesting with multi-array thermoelectric gener-ators,” IEEE Trans. Ind. Electron., vol. 61, pp. 5345–5353, Oct. 2014.

[6] H. Reinisch, M. Wiessflecker, S. Gruber, H. Unterassinger, G. Hofer,M. Klamminger,W. Pribyl, and G. Holweg, “Amultifrequency passivesensing tag with on-chip temperature sensor and off-chip sensor inter-face using EPC HF and UHF RFID technology,” IEEE J. Solid-StateCircuits, vol. 46, pp. 3075–3088, Dec. 2011.

[7] J. Im, S. Wang, S. Ryu, and G. Cho, “A 40 mV transformer-reuse self-startup boost converter with MPPT control for thermoelectric energyharvesting,” IEEE J. Solid-State Circuits, vol. 47, pp. 3055–3067, Dec.2012.

[8] Y. Shih and B. P. Otis, “An inductorless DC-DC converter for en-ergy harvesting with a 1.2- W bandgap-referenced output controller,”IEEE Trans. Circuits Syst. II: Express Briefs, vol. 58, pp. 832–836,Dec. 2011.

[9] R. W. Erickson and D. Maksimovic, Fundamentals of Power Elec-tronics, 2nd ed. Norwell, MA, USA: Kluwer Academic, 2001, pp.13–33.

[10] J. Kim, J. Kim, and C. Kim, “A regulated charge pump with a low-power integrated optimum power point tracking algorithm for indoorsolar energy harvesting,” IEEE Trans. Circuits Syst. II: Express Briefs,vol. 58, pp. 802–806, Dec. 2011.

[11] R. Pagano, M. Baker, and R. E. Radke, “A 0.18- m monolithic Li-ionbattery charger for wireless devices based on partial current sensingand adaptive reference voltage,” IEEE J. Solid-State Circuits, vol. 47,no. 6, pp. 1355–1368, Jun. 2012.

[12] A. Micco, A. Ricciardi, M. Pisco, V. L. Ferrara, L. V. Mercaldo, P. D.Veneri, A. Cutolo, and A. Cusano, “Light trapping efficiency of peri-odic and quasiperiodic back-reflectors for thin film solar cells: A com-parative study,” J. Appl. Phys., vol. 114, pp. 063103–063103-9, Aug.2013.

[13] H. Kim, Y. Min, C. Jeong, K. Kim, C. Kim, and S. Kim, “A 1-mWsolar-energy-harvesting circuit using an adaptive MPPT with a SARand a counter,” IEEE Trans. Circuits Syst. II: Express Briefs, vol. 60,pp. 331–335, Jun. 2013.

[14] K. Ishaque and Z. Salam, “A deterministic particle swarm optimiza-tion maximum power point tracker for photovoltaic system underpartial shading condition,” IEEE Trans. Ind. Electron., vol. 60, pp.3195–3206, Aug. 2013.

[15] T. Esram and P. L. Chapman, “Comparison of photovoltaic array max-imum power point tracking techniques,” IEEE Trans. Energy Convers.,vol. 22, pp. 439–449, Jun. 2007.

[16] H. Shao, C. Tsui, and W. Ki, “The design of a micro power manage-ment system for applications using photovoltaic cells with the max-imum output power control,” IEEE Trans. Very Large Scale Integr.(VLSI) Syst., vol. 17, pp. 1138–1142, Aug. 2009.

[17] Y. K. Ramadass, A. A. Fayed, and A. P. Chandrakasan, “A fully-in-tegrated switched-capacitor step-down DC-DC converter with digitalcapacitance modulation in 45 nm CMOS,” IEEE J. Solid-State Cir-cuits, vol. 45, pp. 2557–2565, Dec. 2010.

[18] S. S. Kudva and R. Harjani, “Fully integrated capacitive DC-DC con-verter with all-digital ripple mitigation technique,” IEEE J. Solid-StateCircuits, vol. 48, no. 8, pp. 1910–1920, Aug. 2013.

[19] J. D. Vos, D. Flandre, and D. Bol, “A sizing methodology for on-chipswitched-capacitor DC/DC converters,” IEEE Trans. Circuits Syst. I:Reg. Papers, vol. 61, no. 5, pp. 1597–1606, May 2014.

[20] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli, “Optimization ofperturb and observe maximum power point tracking method,” IEEETrans. Power Electron., vol. 20, pp. 963–973, Jul. 2005.

[21] L. Salem and Y. Ismail, “Slow-switching-limit loss removal in SCDC-DC converters using adiabatic charging,” in Proc. ICEAC, 2011,pp. 1–4.

[22] W. Xiao, W. G. Dunford, and A. Capel, “A novel modeling method forphotovoltaic cells,” in Proc. IEEE Power Electronics Specialists Conf.,2004, pp. 1950–1956.

[23] M. D. Seeman and S. R. Sanders, “Analysis and optimization ofswitched-capacitor DC-DC converters,” IEEE Trans. Power Electron.,vol. 23, pp. 841–851, Mar. 2008.

[24] R. M. Corless, G. H. Gonnet, D. E. G. Hare, D. J. Jeffrey, and D. E.Knuth, “On the lambertw function,” Adv. Computat. Math., vol. 5, pp.329–359, 1996.

[25] X. Zhang and H. Lee, “An efficiency-enhanced auto-reconfigurable2X/3X SC charge pump for transcutaneous power transmission,” IEEEJ. Solid-State Circuits, vol. 45, no. 9, pp. 1906–1922, Sep. 2010.

[26] B. R. Gregoire, “A compact switched-capacitor regulated chargepump power supply,” IEEE J. Solid-State Circuits, vol. 41, no. 8, pp.1944–1953, Aug. 2006.

[27] P. Chen, X. Zhang, K. Ishida, Y. Okuma, Y. Ryu, M. Takamiya, andT. Sakurai, “An 80 mV startup dual-mode boost converter by charge-pumped pulse generator and threshold voltage tuned oscillator withhot carrier injection,” IEEE J. Solid-State Circuits, vol. 47, no. 11, pp.2554–2562, Nov. 2012.

[28] Y. Qiu, C. Van Liempd, B. Op het Veld, P. G. Blanken, and C. VanHoof, “5 W-to-10 mW input power range inductive boost converterfor indoor photovoltaic energy harvesting with integrated maximumpower point tracking algorithm,” in IEEE ISSCC Dig. Tech. Papers,2011, pp. 118–120.

[29] C. Lauterbach, W. Weber, and D. Romer, “Charge sharing concept andnew clocking scheme for power efficiency and electromagnetic emis-sion improvement of boosted charge pumps,” IEEE J. Solid-State Cir-cuits, vol. 35, no. 5, pp. 719–723, May 2000.

Xiaosen Liu (S'08) received the B.S. degree in elec-trical engineering from Southeast University, Nan-jing, China, and the M.Phil. degree from the HongKong University of Science and Technology, HongKong, in 2008 and 2011. He is currently pursuing thePh.D. degree in electrical engineering at the Analogand Mixed Signal Center (AMSC), Texas A&MUni-versity, College Station, TX, USA.His current research interests include energy har-

vesting, electrosurgical instruments, and dc-dc powerconverters.


Edgar Sánchez-Sinencio (F'92–LF'10) was bornin Mexico City, Mexico. He received the degreein communications and electronic engineering(Professional degree) from the National PolytechnicInstitute of Mexico, Mexico City, Mexico, in 1966,the M.S.E.E. degree from Stanford University, Stan-ford, CA, USA, in 1970, and the Ph.D. degree fromthe University of Illinois at Champaign-Urbana,Urbana, IL, USA, in 1973.He has graduated 59 M.Sc. and 44 Ph.D. students.

He is a coauthor of six books on different topics, suchas RF circuits, low-voltage low-power analog circuits, and neural networks. Heis currently a Distinguished Professor and the TI J. Kilby Chair Professor withTexas A&M University, College Station, TX, USA. His current interests are inthe area of energy harvesting techniques, power management, ultra-low-poweranalog circuits, RF circuits, and medical electronics circuit design.

Prof. Sánchez-Sinencio is a former Editor-in-Chief of IEEE TRANSACTIONSON CIRCUITS AND SYSTEMS II and a former IEEE Circuits and Systems (CAS)Society Vice President–Publications. In November 1995, he was awarded aHonoris Causa Doctorate by the National Institute for Astrophysics, Optics andElectronics, Mexico. This degree was the first honorary degree awarded for mi-croelectronic circuit-design contributions. He was a co-recipient of the 1995Guillemin-Cauer Award for his work on cellular networks. He received theTexas Senate Proclamation #373 for Outstanding Accomplishments in 1996. Hewas also a co-recipient of the 1997 Darlington Award for his work on high-fre-quency filters, the IEEE Circuits and Systems Society Golden Jubilee Medalin 1999, and the prestigious IEEE CAS Society 2008 Technical AchievementAward. He was the IEEE Circuits and Systems Society's Representative to theIEEE Solid-State Circuits Society during 2000–2002. He was a member of theIEEE Solid-State Circuits Society Fellow Award Committee from 2002 to 2004.He has been a member of the ISSCC Analog Committee since 2013. He is aformer Distinguished Lecturer (2012–2013) of the IEEE CAS Society.

an 86% efficiency 12 µw self-sustaining pv energy harvesting

Documents