400 • 2014 IEEE International Solid-State Circuits Conference

ISSCC 2014 / SESSION 23 / ENERGY HARVESTING / 23.4

23.4 Dual-Source Single-Inductor 0.18µm CMOS Charger-Supply with Nested Hysteretic and Adaptive On-Time PWM Control

Suhwan Kim, Gabriel A. Rincón-Mora

Georgia Institute of Technology, Atlanta, GA

A major challenge with emerging microsensors, biomedical implants, and otherportable devices is operational life, because tiny batteries exhaust quickly. Andeven though 1g fuel cells store 5 to 10× more energy than 1g Li-ion batteries,fuel cells supply 10 to 20× less power [1]. This means fuel cells last longer withlight loads and Li-ion batteries output more power across shorter periods.Therefore, when the peak power is far greater than the average power, which, forexample, is typically the case in wireless sensors, a hybrid source can occupyless space than a single source [1]. Still, managing a fuel cell and a battery tosupply a load and recharge the battery, which also acts as an output, with littlespace is difficult. Switched-inductor circuits are appealing in this contextbecause they draw and supply more power with higher efficiency than their linear and switched-capacitor counterparts. Inductors, however, are bulky, andthus microsystems can only rely on one inductor [2-3]. Today, most single-inductor multiple-output systems derive power from one source [3, 4, 6],and therefore, the fuel cell and the battery require considerable space. Systemsthat use two sources either do not manage how much power each source shouldsupply across loading conditions [5], or if they do [2] their efficiency is low. Theadvantage of the prototyped 0.18μm CMOS dual-source single-inductor systempresented here is its less overall volume because it incorporates the functionalintelligence of [2] with much higher efficiency.

Because the 1.1 to 1.3V energy source vES supplies more energy when deliveringconstant power, the prototyped system in Figs. 23.4.1 and 23.4.7 draws constant power PES from vES and supplementary power from the 1.8V powersource vPS to supply an 8mA 0.8V load. When PES exceeds the needs of the loadin PO, the system uses the excess to recharge vPS. Since the hybrid supply usesand switches a 50μH 6×6×2mm3 inductor LO to transfer power between vES, vPS,and the output vO, the purpose of capacitors CIN and CO is to suppress switchingnoise in vES and vO. This way, when PES surpasses PO, the voltage of CO as wellas vO rise above the reference VREF to such an extent that comparator CPM tripsto push the system into the light-load region. CPM pulls the system back into theheavy-mode region when the opposite happens, when PO exceeds PES to pull vO

below the lower hysteretic threshold of CPM.

When lightly loaded, comparator CPLT regulates vO about VREF and LO conducts indiscontinuous conduction mode (DCM) across the period of the 40kHz clockfCLK. More specifically, CPLT senses vO to determine which output: vO or vPS,should receive the energy of LO. For this, SES and SE in Figs. 23.4.1 and 23.4.2first energize LO from vES to ground across τEN’s 1.2μs pulsewidth to raise LO’scurrent iL from zero to 30mA. Afterwards, SES and SE open and vE.OFF in Fig. 23.4.2rises to close SDE and either SO or SPCHG. If comparator CPLT senses vO is belowVREF by 10mV, SO drains LO into vO; otherwise, SPCHG depletes LO into vPS.Comparators CPIOZ and CPIPZ then disengage SO and SPCHG together with SDE whenSO’s and SPCHG’s current nears zero, when LO is close to empty, which happensat 2.7 and 27μs in Fig. 23.4.2. All switches remain open after that until fCLK

initiates another cycle.

CPLT, CPIOZ, and CPIPZ need not operate across the entire 25μs period of fCLK.CPLT, for one, needs to sense vO only at the end of τEN’s 1.2μs pulsewidth. Thisis why fCLK in Fig. 23.4.2 engages CPLT a short delay τD after τEN rises, to be readyby the end of τEN, and disengages CPLT another short delay τD after τEN falls.Similarly, CPIOZ and CPIPZ must sense only when SO and SPCHG conduct iL, soCPLT’s output vLT enables CPIOZ and CPIPZ and CPIOZ’s and CPIPZ’s flip flops disablethem after they detect iL nears zero. Duty-cycling CPLT, CPIOZ, and CPIPZ this wayreduces their power consumption by 90%.

When heavily loaded, LO draws one energy packet from vES and one variablepacket from vPS that transconductor GHV in Figs. 23.4.1 and 23.4.3 controls whenregulating vO about VREF. As with light loads, LO stops conducting after that untilfCLK starts another cycle. Comparator CPHV compares GHV’s slow-moving outputvG against a triangular saw-tooth voltage vSAW to pulsewidth modulate (PWM)how long LO energizes from vPS. For all this, like before, SES and SE first energize

LO from vES to ground across τEN’s 1.2μs pulsewidth to raise LO’s current iL fromzero to 30mA. Afterwards, SES and SE open and SDE and SO close to drain LO intovO. SDE then opens and, if GHV senses that vO still needs power, vSAW starts ramping and SPE closes to energize LO from vPS to vO. When vSAW falls below GHV’svG, SPE opens and SDE closes to deplete LO into vO. SDE and SO then open whenCPIOZ in Fig. 23.4.3 senses that LO’s iL is nearly zero, after which point all otherswitches remain open until the next fCLK cycle.

As Fig. 23.4.4 illustrates, vO ripples at ±2.5mV when lightly loaded with 0.1mAand ±40mV when heavily loaded with 8mA. The ripple is higher at 8mA becausevES and vPS deliver power early in the period and the load slews CO afterwards,when disconnected from LO. Since CPM determines which mode to assert in hysteretic fashion, the system transitions through modes across rising andfalling 0.1 to 8mA load dumps quickly and without ringing oscillations. When theload is light at 0.1 to 1 mA, the fraction of vES power that vO and vPS receive is 62to 73%, as shown in Fig. 23.4.5. The fraction of power vO receives from vES andvPS when heavily loaded with 1 to 8 mA is 62 to 84%. Power-conversion efficiency, ηC, bottoms when the system transitions across 1mA and peaks to73% under hysteretic control below 1mA and 84% under PWM control above 1mA. ηC peaks at two points because switches are smaller in light mode than inheavy mode, thus, conduction and gate-drive losses balance at two load levels.

The key feature of the single-inductor 0.18μm CMOS charger-supply prototypedand validated here is managing two complementary sources with 62 to 84%power conversion efficiency. To achieve this, the system duty cycles circuitblocks, operates the inductor in discontinuous conduction mode, and employshysteretic and PWM control schemes to regulate the output in and across lightand heavy modes. The challenge with single-source systems when lightly loadedover extended periods and pulsed periodically with heavy loads, as in the caseof wireless sensors, is that oversizing a fuel cell to output more power or a Li-ion battery to last longer demands more space than an efficient hybrid. Tosustain a 0.1 to 10mW load for one month, for example, [4] and [6] in the tableof Fig. 23.4.6 require a 1g fuel cell to supply 10mW or a 0.45g or 0.43g Li-ionbattery to last one month. Since the circuit proposed in [5] cannot adjust howmuch power each source should supply according to the load, [5] needs a 1gfuel cell or a 0.43g Li-ion battery. [2] can manage a fuel cell and a Li-ion batteryaccording to the load, but the cost of intelligence, robustness, and accuracy isunfortunately efficiency, so this hybrid system demands more space than [4-6].The dual-source single-inductor charger–supply presented here, however,requires a 0.1g fuel cell and a 0.05g Li-ion battery, which at 0.15g combinedresults in 65% less weight than that of the smallest counterpart.

Acknowledgements:The authors thank Pooya Forghani, Paul Emerson, and Texas Instruments fortheir support and for fabricating the prototyped IC.

References:[1] M. Chen, J.P. Vogt, and G.A. Rincón-Mora, “Design Methodology of a HybridMicro-Scale Fuel Cell-Thin-Film Lithium Ion Source,” IEEE Int’l Midwest Symp.on Circuits and Systems, pp. 674-677, Aug. 2007. [2] S. Kim and G.A. Rincón-Mora, “Single-Inductor Dual-Input Dual-OutputBuck-Boost Fuel Cell-Li Ion Charging DC-DC Converter,” ISSCC Dig. Tech.Papers, pp. 444-445, Feb. 2009.[3] D. Ma, W.H. Ki, C.Y. Tsui, and P.K.T. Mok, “Single-Inductor Multiple-OutputSwitching Converters with Time-Multiplexing Control in DiscontinuousConduction Mode,” IEEE J. Solid-State Circuits, vol. 38, no. 1, pp. 89-100, Jan.2003.[4] M.H. Huang and K.H. Chen, “Single-Inductor Multi-Output (SIMO) DC-DCConverters with High Light-Load Efficiency and Minimized Cross-Regulation forPortable Devices,” IEEE J. Solid-State Circuits, vol. 44, no. 4, pp. 1099-1111,Apr. 2009. [5] K.W.R. Chew, Z. Sun, H. Tang, and L. Siek, “A 400nW Single-Inductor Dual-Input-Tri-Output DC-DC Buck-Boost Converter with Maximum Power PointTracking for Indoor Photovoltaic Energy Harvesting,” ISSCC Dig. Tech. Papers,pp.68-69, Feb. 2013.[6] Y. Qiu, C.V. Liempd, B. Op het Veld, P.G. Blanken, and C.V. Hoof, “5μW-to-10mW Input Power Range Inductive Boost Converter for Indoor PhotovoltaicEnergy Harvesting with Integrated Maximum Power Point Tracking Algorithm,”ISSCC Dig. Tech. Papers, pp. 118-119, Feb. 2011.

978-1-4799-0920-9/14/$31.00 ©2014 IEEE

401DIGEST OF TECHNICAL PAPERS •

ISSCC 2014 / February 12, 2014 / 10:15 AM

Figure 23.4.1: Dual-source single-inductor charger-supply IC. Figure 23.4.2: Light-load mode control and waveforms.

Figure 23.4.3: Heavy-load mode control and waveforms.

Figure 23.4.5: Simulated and measured efficiency of the prototype IC. Figure 23.4.6: Performance summary and comparison.

Figure 23.4.4: Mode transitions in rising and falling load dump responses.

23

• 2014 IEEE International Solid-State Circuits Conference 978-1-4799-0920-9/14/$31.00 ©2014 IEEE

ISSCC 2014 PAPER CONTINUATIONS

Figure 23.4.7: Die micrograph and printed circuit board.