Battery-Assisted Battery Charger with Maximum Power Point Tracking for Thermoelectric Generator: Concept and Experimental Proof

Abstract

This paper proposes a concept of battery-assisted battery charger with maximum power point tracking for DC energy transducer such as thermoelectric generator and photo voltaic generator, and shows experimental results to prove the concept. The DC energy transducer is connected in series with a battery to increase the voltage. The plus terminal for the DC energy transducer is connected with the input terminal of a DC-DC buck converter, whereas the battery is connected with the output terminal of the converter. Thus, the current is boosted from the input to the output. When the net current to the battery is positive, the system works as a battery charger. To extract the as much power from the DC energy transducer as possible for high charging efficiency, maximum power point tracking is introduced. The converter was designed in 180 nm 3V CMOS with a silicon area of 1.05 mm². The concept was experimentally proven by varying the reference voltages to control the input voltage. An all-solid-state battery was charged up from 2.2 V to 2.3 V in two hours by the converter with a flexible thermoelectric generator which had an open-circuit voltage of 0.6 V.

1. Introduction

Internet of Things (IoT) sensing modules communicate with one another or cloud servers to collect the information around them, and to keep society safe with little or no human intervention [1,2]. However, there is a need to replace the battery used in the sensor modules. The cost of replacing old batteries with new ones has increased significantly as many sensor modules are distributed globally. To decrease the overall cost of sensor networks, technology is needed to eliminate battery replacement. Energy harvesting is a key technology for eliminating battery replacement. One solution is the use of an energy transducer (ET) to convert the surrounding energy, such as thermal flow and light, into electric power for sensor modules without any battery [3,4].

Maximum power point tracking (MPPT) is a key design technique for the converter to extract as much power as possible from a thermoelectric generator (TEG) [5,6,7,8] and a photo voltaic generator (PVG) [9,10]. Fully integrated charge-pump circuits as converter have three main design parameters of clock frequency f, number of stages N and stage capacitor C. MPPT utilizes one or more design parameters to maximize the input or output power of the charge pumps: N and f (two-dimensional MPPT) [5,6], C, N, f (three-dimensional MPPT) [7], and one-dimensional MPPT [8]. There are two categories for MPPT, direct and indirect methods [9]. The direct method varies the output voltage and the current of PV to determine the maximum power point. The indirect method estimates MPP based on the measured results of the PV generator’s voltage and current, the irradiance, or using empiric data, via mathematical expressions of numerical approximations. Because there is no battery, there is no need to replace batteries. However, when the modules require operation more often than the cycle time of the surrounding energy, a battery is required as a backup [11,12,13,14,15,16]. In [11], a charger and battery management IC with 330 nA quiescent current was presented. The IC can cold start from 330 mV and 5 μW of input power. The charger achieves an efficiency greater than 80% at single cell solar voltages of 0.5 V. In [12], a converter was proposed to charge battery and to drive the load simultaneously. For an input power of 500 nW, the proposed chip achieved an efficiency of 82%, including the control circuit overhead, while charging the energy storage device at 3 V from 0.5 V input. In buck mode, it achieved a peak efficiency of 87% and maintained an efficiency greater than 80% for an output power of 50 nW–1 μW, with an input voltage of 3 V and an output voltage of 1 V. In [13], a battery charger for electrostatic vibration energy transducer was presented. In [14,15], current-controlled chargers were presented for Li-ion batteries. To preferably regulate the charging current and decrease circuit complexity for parallel charging, a battery charger with variable charging current and automatic voltage-compensation controls was presented in [14]. The main architecture adopted was a two-loop current-mode control in the constant current and the constant voltage stages in [15]. Trickle-current mode provided complete battery charging process to protect the battery. In [16], a hybrid TEG—battery power supply system was proposed to achieve replacement-free batteries in sensor modules. A TEG was connected in series with the battery to increase the input voltage of a single DC-DC buck converter to drive a sensor IC, and to return the majority of the battery power to the battery in every cycle. Discontinuous mode (DCM) was applied to improve the power conversion efficiency at light loads [17]. The high side was enabled when the voltage at the load was below a target voltage. The activated period was controlled with a predetermined pulse width. The low side was enabled as soon as the high side was turned off. The low side was deactivated when the drain-to-source voltage returned zero. Even though the battery charging operation was validated with experiments, it had no MPPT capability.

In this paper, a battery charger is proposed based on the hybrid TEG—battery power supply system. The contributions to industrial electronics and energy harvesting technologies are as follows:

(1)

The proposal of the battery-assisted battery charger. It allows standard CMOS to design the battery charger even with the TEG whose open-circuit voltage is as low as 0.6 V, because the input voltage of the battery charge is the sum of the battery voltage and the TEG output voltage.

(2)

The proposal and experimental proof of the MPPT for the battery-assisted battery charger. The theoretical maximum input power extracted from TEG is shown and has been experimentally proven to charge the battery with maximum power.

This paper is organized as follows. In Section 2, a battery-assisted battery charger with MPPT is proposed. The MPPT circuit is shown with a flow chart for the converter operation. Block diagram of the converter chip is discussed in detail. In Section 3, measured results are presented. The input voltage of the converter is skewed with reference voltages varied to determine the MPP for TEG. In Section 4, future works are discussed. Section 5 summarizes the paper.

2. Battery-Assisted Battery Charger with Maximum Power Point Tracking

Figure 1a shows a battery, which will be connected to a sensor IC to supply power. Figure 1b illustrates a converter system to supply power to a sensor IC, and to charge the battery in every operation period when TEG generates sufficiently high power [16]. Figure 1c proposes a battery system with a battery-assisted battery charger. The battery is connected to a sensor IC as usual, but the battery can have longer lifetime than normal batteries with a DC energy transducer and a converter. To operate the battery charger, it requires P_ASSIST from the battery. As long as energy transducer generates sufficiently large power, the output power from the converter, P_GAIN, can be larger than P_ASSIST. When the net power to the battery is positive, the battery is charged up.

To extract as much power from the DC energy transducer as possible for high charging efficiency, maximum power point tracking (MPPT) is required. When the DC energy transducer is a thermoelectric generator (TEG), it is modeled with a DC voltage source V_EH and an output resistance R_EH. Buck converter needs to operate the output voltage of TEG V_EHO around V_EH/2 to maximize the output power of TEG P_EH [18,19,20]. When the DC energy transducer is a photo voltaic generator (PVG), it is modeled with a DC current source I_EH, a diode to determine the maximum output voltage V_PV, and an output resistance R_EH [9,10]. The maximum power point is found to be somewhere in V_OP > V_EH/2 depending on the model parameters [9].

Figure 2a shows a target operation for the input voltage V_in of the battery-assisted battery charger.

There are two operation modes: T_refresh to refresh the reference voltage to control V_in, and T_MPPT to operate the converter around the maximum power point. T_refresh has two period: T_wait to wait V_in to be saturated at V_{in_open}, and T_dec to sample and hold V_eh. T_MPPT has two operation periods: T_sus and T_res to suspend or resume the converter operation, respectively. T_MPPT should be shorter than that of drift in temperature difference in case of TEG and luminous intensity in case of PVG in order to keep the power system at MPP, whereas much longer T_refresh is needed to have sufficient power conversion efficiency. If T_MPPT is as short as T_refresh, the average power conversion efficiency can be as low as 50% at most.

Figure 3 shows the flow chart of the converter control. The algorism starts when ET starts outputting power (S1). When ET outputs power, V_in increases. Power-on reset circuit (POR) outputs a signal PONRST (S2) to enable a counter to control T_wait, T_dec, T_refresh and T_MPP (S3). Until T_wait, V_in is supposed to be saturated at the open-circuit voltage. Based on the type of ET, TEG or PV, the reference voltages V_in+ and V_in− are determined via the A/D and D/A converters within T_dec (S4, S5). After the target voltage for V_in is refreshed, the converter is enabled to operate in MPPT in T_MPPT (S6, S7). After T_MPPT, the converter is disabled (S8) for the next refresh period (S3–S8).

Figure 4a,b, respectively show the configuration of the proposed battery charger and the block diagram of the converter chip. The aim of this paper is to propose the design of a battery-assisted battery charger with MPPT, and provide its experimental proof. The experimental circuit should have the capability of variables V_in+ and V_in− to validate the optimum V_in for maximizing the battery charging power. Thus, the circuit was designed to control V_in+ and V_in− externally. The design concept for the reference voltage generator will be discussed in Section 4. In Figure 4b, V_in is the power supply pin for the control circuit, whereas V_{in_pmos} is the pin for the power switch to flow the current into the inductor. They are connected with TEG at the top level. Those two pins can be merged into a single pin, but were kept separate in this test chip in order to measure the current separately. V_d is connected with an inductor. “CONV_MPPT” is the power switch to connect V_{in_pmos} and V_d in the first period, and V_d and ground in the second period. Thus, V_d toggles from V_in to ground and vice versa. “CURRENT MIRROR_CKT” is a voltage-controlled current-generator. For this test, it receives input voltages at CRT_all and CRT_dly to generate reference currents for several differential amplifiers and delay circuits, respectively. One can tune the input voltages so that the comparators function well with minimal currents, or the conduction time for the high side of the power switches is adjusted to have a target peak inductor current. Once optimum input voltages are determined, those two pins can be removed in the following design. “PONRST” is a power-on reset circuit to detect whether V_in becomes high enough. The output signal of “PONRST” wakes a bandgap reference “BGR” to generate a reference voltage. “BGRO” is a monitoring circuit for the reference voltage. With high EN_BGRO, the reference voltage is output from a pin BGRO. “OSC” controls the pulse width for the refresh and MPPT periods. Once V_in is high enough, “OSC” starts working. “MPPT_CKT” inputs V_in/2 and two reference voltages V_in+/2 and V_in−/2, and outputs an enabling signal to “PGNG_CKT”. Once V_in reaches V_in+, the power switches resume the switching operation until V_in reaches V_in−; switching operation gradually pulls down V_in. Once V_in reaches V_in−, the power switches suspend the switching operation until V_in reaches V_in+. V_in gradually increases with TEG. This suspend/resume operation is repeated during T_MPPT. “MONITOR_ALL” outputs the buffered signals of the gate signals for the power switches, NG and PG, with a low OEB. In Figure 4b, each of the critical building blocks is labelled by a specific figure number such as Figure 5a,b, Figure 6, Figure 7a,b or Figure 8a.

Figure 5a illustrates the voltage-controlled current-generator “CURRENT MIRROR_CKT”. It receives an input voltage at CRT_IN to draw a reference current via R1, and outputs voltages V_gn and V_gp to control the current sources in other comparators, oscillator, and bandgap. Figure 5b shows the power-on reset circuit “PONRST”. The comparator compares V₋ = V_in − |V_tp| and V₊ = R3/(R3 + R2)V_in. When V_in becomes higher than a point where V₋ = V₊, PONRST becomes low. With this low PONRST, the bandgap reference (BGR) starts working.

Figure 6 illustrates the “OSC” which was introduced in Figure 4. Figure 6a shows a binary counter composed of a ring oscillator (RINGOSC) and several D-Flip-Flops (D_FFs). Figure 6b illustrates a pulse generator to define T_refresh and T_MPPT. Figure 6c shows waveforms of the outputs of D_FFs and a control signal enb, which is generated via LATCH, shown in Figure 6b, as Q. The colored dashed lines indicate that those signals are synchronized. The latch gets reset with a power-on reset signal to “enable”. Q_set and Q_reset of Figure 6b are connected with signals C4 and OSCb, respectively. One can vary T_refresh and T_MPPT by connecting different outputs of D_FFs with Q_set and Q_reset.

Figure 7a shows the RES/SUS signal generator. RES increases, i.e., the latch is set, when the converter enters into MPPT or when V_in hits V_in−. SUS increases, i.e., the latch is reset, when V_in hits V_in+. Thus, V_in is controlled within V_in+ and V_in−. Figure 7b shows the PG/NG pulse generator. PG and NG are the gate signals of pull-up and pull-down drivers, as shown in Figure 8a. The waveform is shown in Figure 8b. During RES period, the converter operates in a boundary conduction mode, which requires a simple control circuit. PG decrease with an increase in RES. PG returns to a high rate after a predetermined pulse width is determined by a delay cell DELAY_ALL. When the PG is high, NG increases. There is a timing that both pull-up and pull-down drivers turn off in order to ensure that there is no direct current from power supply V_DD, which is connected with V_in at the top-level cell, to the ground. NG decreases when Vd reaches 0 V, i.e., zero-volt switching. With a reduction in NG, PG reduces. The above sequence is repeated until SUS signal increases. When SUS increases, NG is forced to reduce regardless of the present state. After the following PG toggle, PF is stuck to increase. Hence, both the drivers turn off until the next RES signal. Thus, the battery is charged up with the inductor current.

Figure 9 shows the current loops in the battery charger. The loop in light blue is continuous during T_MPPT. When V_in+ − V_in− is much smaller than V_in+, i.e., V_in is regarded to be constant; the current value is also constant. The current direction is from the battery to the converter. The loops in red during Tres is direct from the converter to the battery.

Table 1. Comparison of this work with the previous one.

	Previous Work [16]	This Work
Operation mode	Discontinuous conduction mode	Boundary conduction mode (during RES period)
Outputs of the converter	Load and battery	Battery-only
Parameter to be controlled	VLOAD to be controlled to the target voltage	Vin− < Vin < Vin+
Charged battery power	– Depending on PLOAD – Converter output power split into the load and the battery	To be maximized

compares the differences of the proposed battery-assisted battery charger from the previous work [16].

3. Experiments

In this section, we report on how the circuit parameters have been determined, and present the measured results in the case of an equivalent circuit model for TEG, a flexible TEG [21], with a mimic heat pipe. The assumed flexible TEG has similar electrical parameters with the ones used in [8,22].

3.1. Determination of the Circuit Parameters

The assumed TEG has an open-circuit voltage of 0.6–1.2 V and an output resistance of 0.6–1.2 kΩ. As a result, TEG can output power of several hundred micro-watt. A potentially maximum peak input current to the converter is assumed to be 3 mA at most, and 180 nm 3 V CMOS was available. Assuming that maximum Vds of the power MOSFETs is as low as 20 mV, the total channel width of PMOS and NMOS was selected to be 2000 μm and 1000 μm, respectively. The channel length of PMOS and NMOS was set at a minimum of 0.3 μm and 0.36 μm, respectively. Based on a previous design [16], the minimum converter cycle time was assumed to be 10 μs. Thus, the inductance was determined with a rough estimate of V_EH × T_p/I_peak~1 V × 10 μs/3 mA~3 mH. The 3 V CMOS allows the maximum input voltage of 3.6 V to avoid any overstress issue. Thus, it is assumed that the nine maximum battery voltage is 2.3 V, and the maximum V_EH is 1.2 V in this demonstration. The size of a supply capacitor for V_in (C_in) may affect converter operation because a too small C_in only allows a few suspend/resume cycles, which may affect the power conversion efficiency. Thus, C_in of 0.67 μF and 3.4 μF were tested. The components’ values are summarized in

Table 2. Components’ values of measured system.

Parameter		Symbol	Value
Input power supply capacitor		Cin [μF]	0.67, 3.4
Inductor		L [mH]	3.0
Battery		VBAT [V]	1.6, 2.3
Equivalent circuit of TEG	Open circuit voltage	VET [V]	0.6, 1.2
	Equivalent output resistance	RET [kΩ]	1.2

.

Figure 10 shows a die photo of the converter chip fabricated in 180 nm 3V standard CMOS. It does not need low-Vt transistors because V_in is the sum of V_BAT and V_EHO. Poly resistors used for BGR and voltage divider occupies 70% of an entire area of 1.05 mm² for low-power operation.

3.2. Equivalent Circuit Model for TEG

Figure 11a,b shows the measured waveform for V_in in yellow, V_d in light blue, PG in red, and I_L in green under the condition of C_in = 0.67 μF, V_BAT = 1.6 V, V_EH = 1.2 V, V_in+/2 = 1.13 V, V_in−/2 = 1.08 V. T_refresh and T_MPPT are shown in the arrows. V_in was controlled at 2.27–2.15 V. T_res starts with PG low. The inductor current linearly increased in a delay of 25 μs and had a peak current of 4.0 mA. Thus, V_in is pulled down. With PG (and NG) high, the inductor current linearly decreased until V_d hit the ground. Because V_in was separate from the power switches and the output resistance of the equivalent TEG model was relatively high, it remained the same. Three cycles pulled V_in down to V_in− in case of a small C_in of 0.67 μF. The MPPT controller disabled converter operation until V_in was pulled up to V_in+.

Let us see if three cycles of the converter operation were matched with the expected one. Assuming that the variation in V_in is much smaller than V_in, i.e., the inductor current increases linearly, the following equations are formulated, where I_PK is the peak inductor current and T_H (T_L) is the active duration of the high side PMOS (the low side NMOS).

$$L\frac{I_P}{T_H}=V_{EHO}$$

(1)

$$L\frac{I_P}{T_L}=V_{BAT}$$

(2)

Using (1), the average high side current I_P over one converter operation cycle T_C can be obtained:

$$I_P=\frac{I_{PK} T_H }{2T_C}=\frac{T_{H}2 V_{EHO} }{2L{T}_C}$$

(3)

The output current of TEG I_EH is calculated using

$$I_{EH}=\frac{V_{EHO} }{R_{EH}}$$

(4)

A voltage drop in V_in per converter operation cycle, ΔV_in, can be estimated using (5).

$$C_i\frac{\mathsf{\Delta }{V}_{in} }{T_C}=I_P-I_{EH}$$

(5)

When V_EHO is 0.6 V and T_H is 22μs, ΔV_in is calculated to be 36 mV with (1)–(5). In this experimental setup, V_in+ −V_in− was set to be 0.12 V. As a result, a required number of converter operation cycle is expected to be three, which is matched with the measured result.

Average effective output power of the ET, P_eh, defined by I_eh × V_EHO, and average effective input power to the battery P_bat were measured with different sets of V_in+ and V_in−, as shown in Figure 12a. Battery charger efficiency can be defined by P_bat/P_{eh_peak}, where P_{eh_peak} indicates the peak value of P_eh, as shown in Figure 12b. Approximate curves are all the second order functions of V_eho/V_eh. All curves have their peak values at V_eho/V_eh of 0.5–0.55 as expected. The maximum difference between SPICE results and the measured ones in the battery charger efficiency was 10% because measured power was 5% for P_eh when SPICE result was higher, whereas measured power was 5% for P_bat when SPICE result was lower. The root cause of those discrepancies needs to be identified, which will be a focus of future works.

The system operated at a low V_eh of 0.6 V, as shown in Figure 13 and Figure 14. Figure 13a and Figure 14a are waveforms of SPICE, and Figure 13b and Figure 14b are measured waveforms. Under the condition of C_in = 3.4 μF, V_BAT = 1.6 V, V_EH = 0.6 V, V_in+/2 = 1.0 V, and V_in−/2 = 0.95 V. The measured waveforms were in good agreement with the SPICE results.

Figure 15a, b respectively show power and efficiency when V_eh = 0.6 V. Although the efficiency was as low as 50–60%, the battery was charged with a power of 40–50 μW.

3.3. Flexible TEG with a Mimic Heat Pipe

Figure 16 shows two pieces of flexible TEG arch [21] which can be attached to a mimic heat pipe. The inner diameter is heated, whereas the outer one is at room temperature. A unit TEG generates an open-circuit voltage of 0.6 V, and has an output resistance of 0.6 kΩ when the inner diameter is heated at 100 °C. Such high temperature is available in chemical factories [23]. One application is the detection of a symptom that the temperature is increasing beyond the threshold. The sensor can issue an alarm before a terrible accident occurs. An all-solid-state battery of 2.3 V suitable for energy harvesting applications [24] was used to confirm that the circuit system can charge a battery even with TEG whose maximum attainable output power is 150 μW.

Figure 17 shows that the converter system with a real TEG and a rechargeable battery charged up the battery with a maximum charging efficiency of 38%. Higher battery voltage tends to reduce power efficiency [16].

The battery-assisted battery charger was left working at a V_eho of 0.3 V in 5 h. The battery voltage was monitored periodically. Figure 18 shows the battery charging characteristics. The circles indicate measured points. The battery voltage increased from an initial voltage of 2.21 V to 2.28 V after one hour. The battery voltage was saturated at 2.3 V. Thus, the battery-assisted battery charger was confirmed to charge the battery with TEG.

4. Future Work

Even though the concept of the battery-assisted battery charger was validated with the experimental results, the following sub-blocks need to be added in the future. The first one is a reference voltage generator, as shown in Figure 19. In this research, V_in+ and V_in− were provided externally to validate MPPT operation. In production-level converters, they need to be generated internally based on the open-circuit voltage V_{in_open} monitored in the refresh period, as shown in Figure 19a. Figure 19b shows the timing chart. Using step 4, which is presented in Figure 3, the capacitors C1 and C2 are equalized at V_{in_open} (=V_BAT + V_EHO) and V_BAT, respectively, with high Φ1. At step 4, Φ1 and Φ2 are set to be low and high, respectively. When the capacitor sizes of C1 and C2 are equal, the capacitor voltage becomes V_BAT + V_EHO/2, which is the current MPPT voltage V_MPPT. A/D converter outputs an updated digital value associated with V_MPPT by T3. D/A converter latches the digital value D<4:0> at T3 and outputs V_in+ and V_in− at the appropriate values of V_MPPT + ΔV/2 and V_MPPT − ΔV/2, respectively, where ΔV is a predetermined ripple voltage in V_in, e.g., 50 mV.

The second additional sub-block is monitoring circuits for V_{in_open} and V_BAT to suspend the converter operation when one of them is too low to reliably operate the converter. Over-charging protection circuit is also needed to prevent the battery from over-charging.

5. Conclusions

The battery-assisted battery charger was proposed for energy harvesting for a thermoelectric generator. The operation cycle was composed of refresh and MPPT periods. In the refresh period, the open-circuit voltage at V_in, i.e., V_BAT + V_EH, was measured to determine the V_MPPT of V_BAT + V_EH/2 using the switched-capacitor analog calculator. In the MPPT period, the converter was controlled to keep the V_in around V_MPPT between V_in+ and V_in−. The MPPT period was composed of suspend and resume periods. In the suspend period, switching operation of the converter was disabled, and so V_in gradually increased toward V_in+. Once V_in hits V_in+, switching operation was resumed, and so V_in gradually decreased toward V_in−. In the resume period, the converter operated in a boundary conduction mode. The input power to the converter is the sum of the power of the battery and power of the TEG. As a result, the converter can charge the battery with the TEG.

The converter circuit was fabricated in 180 nm 3 V standard CMOS with 1.05 mm². The input reference voltages were varied to validate the maximum power tracking point, which was confirmed to be located at 50–60% of the open-circuit voltage of TEG. The battery-assisted battery charger with a flexible TEG and an all-solid-state battery charged the battery from 2.21 V to 2.28 V in one hour, which confirmed the functionality of the proposed battery charger. Boundary conduction mode (BCM) was applied to the converter design. In future work, such a battery could be designed in a continuous conduction mode and compared with BCM.