Efficient Power Conversion for Ultra Low Voltage Micro ...

Efficient Power Conversion for Ultra Low Voltage Micro Scale Energy Transducers

Chao Lu, Sang Phill Park, Vijay Raghunathan, Kaushik Roy School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907

<lu43, sppark, vr, kaushik>@purdue.edu

Abstract – Energy harvesting has emerged as a feasible and attractive option to improve battery lifetime in micro-scale electronic systems such as biomedical implants and wireless sensor nodes. A key challenge in designing micro-scale energy harvesting systems is that miniature energy transducers (e.g., photovoltaic cells, thermo-electric generators, and fuel cells) output very low voltages (0-0.4V). Therefore, a fully on-chip power converter (usually based on a charge pump) is used to boost the output voltage of the energy transducer and transfer charge into an energy buffer for storage. However, the charge transfer capability of widely used linear charge pump based power converters degrades when used with ultra-low voltage energy transducers. This paper presents the design of a new tree topology charge pump that has a reduced charge sharing time, leading to an improved charge transfer capability. The proposed design has been implemented using 65nm technology and circuit simulations demonstrate that the proposed design results in an increase of up to 30% in harvested power compared to existing linear charge pumps.

I. INTRODUCTION Rapid advances in nanoscale integration have resulted in a

new class of miniaturized electronic systems (e.g., smart dust sensors [1], biomedical implants [2]) that enable new application domains. Despite the stringent constraints on size (and hence battery capacity), these systems are often required to operate for several months to years without the need for battery replacement, because frequent battery replacement may be infeasible (e.g., for biomedical implants) or prohibitively expensive (e.g., in a large sensor network). Environmental energy harvesting represents an attractive approach to alleviate the energy supply challenge in these systems and has the potential to result in self-powered, perpetual system operation [3] [4].

While the basic idea of environmental energy harvesting has been extensively explored and applied at the macro-scale in the context of large systems such as solar farms, windmills, and hydro-generators, designing a micro-scale energy harvesting subsystem for miniaturized electronic devices involves several new challenges. First, the form-factor constraint in these systems mandates the use of very small energy transducers (a few cm2 or cm3). As a result, the output voltage of the transducer is very low, often far less than 1V. For example, as we demonstrate in Section II, miniature single junction photovoltaic (PV) cells and thermo-electric generators (TEG) produce voltages in the range of 0.2-0.6V. This is also true for other energy sources such as silicon based micro-fuel cells [6]. Extracting energy from such an ultra-low voltage source requires judicious circuit design [7]. Second,

the maximum power output of these micro-scale transducers is extremely small, often only a few milliwatts. Therefore, the harvesting subsystem should be carefully designed to extract as much power as possible from the transducer and transfer it to the electronic system. Note that while it is possible to stack multiple energy transducer modules (i.e., connect them in series or parallel) to generate higher output voltage and current, such an option is often not viable in micro-scale systems due to size, cost, and packaging considerations [7].

A key component of any micro-scale energy harvesting subsystem, and the focus of this paper, is the power converter that boosts the output voltage of the energy transducer to a suitable level that enables energy storage in a rechargeable battery or an ultra-capacitor. The power converter can be implemented using either an inductive boost converter or a charge pump. Boost converters require an external bulky inductor, which increases system cost and size. Therefore, charge pumps, which consist only of capacitors and MOS switches, are a preferred alternative for fully on-chip implementation. The first proposed charge pump circuit was a Dickson charge pump that had a linear topology [8]. Recently N-stage linear charge pumps have been extensively studied and optimized [9-11]. As we will show later, the performance of such linear topology charge pumps degrades substantially when used with an ultra-low voltage energy transducer. In order to improve power conversion efficiency, cross-coupled charge pumps have also been proposed [12] [13]. However, when the input voltage is lower than the threshold voltage of transistors, the cross-coupled topology fails to function normally. In [14], an exponential topology charge pump was proposed to extract energy from ultra-low voltage sources. However, the charge transfer capability of this topology is still impeded by the first stage.

The paper makes the following contributions: (a) We characterize two commercially available micro-scale energy transducers, a PV cell from Solar World Inc. and a TEG from Micropelt Inc. (b) Based on the characterization, we present the design of an optimized power converter for such ultra-low voltage (i.e., 0-0.4V) energy transducers. The proposed power converter is based on a tree topology charge pump with improved charge transfer capability. The operation of the proposed design is analyzed in detail and compared with that of linear topology charge pump designs. (c) We have implemented the proposed design using IBM 65nm technology and present simulation results to quantify its benefits. Our results demonstrate that the proposed power converter results in an increase in harvested power by up to 30% compared to linear topology charge pump based designs.

978-3-9810801-6-2/DATE10 © 2010 EDAA

The remainder of this paper is organized as follows. Section II discusses the power characterization of two miniature energy transducers. Section III presents the design of a power converter that is optimized for use with ultra-low voltage energy transducers. The operation of the proposed power converter is analyzed in detail and compared to an optimized linear charge pump. Section IV describes the implementation of the proposed power converter and presents simulation results to demonstrate its benefits, while Section V concludes the paper.

II. ENERGY TRANSDUCER CHARACTERISTICS

A. Photovoltaic Cell Figure 1 shows the electrical model of a PV cell [11] [15].

IPH, SC is the generated photocurrent, which varies with light irradiance, RS is the parasitic serial resistance, RP is the equivalent shunt resistance, Io and VPH are the output current and voltage from the PV cell.

Figure 1. Photograph of a miniature PV cell and its electrical model

Figure 2. I-V characteristic and output power of the Solar World PV cell

We conducted experiments on a miniature non-stacked

PV cell (Model 1-100 from Solar World Inc. [16]) to measure its I-V characteristics. To ensure reproducibility of the experimental setup, the PV cell was illuminated by a 40-Watt lamp and the distance between the lamp and the PV cell was adjusted to represent variable light conditions. Various resistive loads were connected to the PV cell and the measured output voltage and current were recorded. The I-V curve of the PV cell is plotted in Figure 2. As can be seen, the PV cell behaves as a voltage limited current source, confirming the electrical model shown in Figure 1. Further, as seen in Figure 2, the output voltage of the PV cell is very low (less than 0.42V) even for high irradiance. Figure 2 also plots the output power of the PV cell as a function of the output voltage and shows that, for any given light intensity, there is an optimal PV cell voltage where the output power is maximized. Further, it can be seen that the power output from the transducer is very small (mostly less than 1mW).

B. Thermoelectric Generator Thermoelectric generators (TEG) typically consist of

multiple couples of p-type and n-type thermoelectric legs, which can output electrical energy by employing the temperature gradient between the hot surface (human body) and the cold surface (ambient air) [5]. Micropelt [17] MPG-D751 is a good example of a miniature TEG device. The current and voltage values for different ΔT across it were obtained using the simulation tool provided by the manufacturer [18]. These are plotted in Figure 3. The output power varies as a function of output voltage for different ΔT. As can be seen, the output voltage of the TEG is in the range of 0-0.6V and the maximum power output is around 0.25mW.

Figure 3. Simulation results of TEG MPG-D751 (ΔT=1~4K)

III. SYSTEM DESCRIPTION AND ANALYSIS

Figure 4 shows the block diagram of the proposed energy harvesting system. It consists of four main blocks: the energy scavenger, the power converter, the energy buffer and the application unit.

Figure 4. Proposed energy harvesting system diagram

The energy scavenger harvests energy from the environment and produces unregulated low voltage. The energy scavenger can be a non-stacked photovoltaic cell, thermo generator, fuel cell or a combination of them. The unregulated low voltage of the energy scavenger block is supplied to the power converter unit, where a high conversion ratio charge pump steps up the voltage and transfers the harvested charge into an energy buffer. The control unit provides the desired switching frequency with enough capability to drive the charge pump. The clock generator inside the control unit can be implemented as a ring oscillator or an analog voltage controlled oscillator. The non-overlapping block provides non-overlapped clock signals (Φ and ΦB) to avoid the reverse current in the charge pump. Energy previously stored in the energy buffer enables the

initial start-up of the charge pump. The energy buffer also acts as the power supply for the control unit. In the next subsections we briefly discuss existing linear charge pump topologies and present an enhanced charge pump design based on tree topology.

A. Linear Charge Pump An optimized 3-stage 4× linear charge pump that has an

ideal voltage step up ratio of 4 is shown in Figure 5. It is used to transfer the harvested current from an ultra low voltage energy scavenger to an energy buffer. During each clock cycle, the charge from the previous stage is stored on the capacitor and is then transferred to the subsequent stage. At the final stage, the harvested charge from the energy scavenger is dumped into the energy buffer for storage.

Figure 5. 3-stage linear charge pump circuit

The average output current from a 3-stage linear charge pump is given by [10], [11]

The above equation is derived based on the assumption that complete charge sharing is guaranteed within each clock cycle. It reveals a linear relationship between the charge pump output current and switching frequency fclk when the capacitor (C), scavenger voltage (Vsource) and energy buffer voltage (Vbuffer) do not change. When the clock frequency is low, the output current varies linearly with the applied clock frequency. However, when the frequency increases beyond a certain value, the charge sharing becomes incomplete. Hence, the linear relationship between output current and switching frequency does not hold beyond the knee frequency due to gradual saturation of the output current. Figure 6 illustrates the variation of charge pump output current with the switching frequency. The interface frequency between complete or incomplete charge sharing is marked as knee frequency.

Figure 6. Charge pump output current vs. the applied switching frequency

The knee frequency is an intrinsic property of a charge pump and is determined by the time constant of charge sharing paths within it. The MOS transistors in the charge pump can be modeled as a voltage-controlled switch with non-negligible on-resistance which can be expressed as

Note that the linear charge pump in Figure 5 has the minimum on-resistance because the gate voltage (VG) of each switch utilizes the feeding voltage back from the energy buffer. As a consequence, the charge pump in Figure 5 is an optimized topology that has superior charge transfer capability than other linear ones.

From (2), it can be inferred that reduction in threshold voltage (VTH) will result in smaller on-resistance as well as enhanced charge transfer capability. However, leakage current increases exponentially with the scaling of threshold voltage. In order to avoid huge leakage current, transistor threshold voltage cannot be scaled down considerably. Hence, the challenge of limited charge transfer capability in the linear charge pump will not be alleviated in future scaled technologies.

Equation (2) also reveals that the conduction resistance Ron diminishes with the decrease of VS+VD. There are three charge sharing paths inside the 3-stage linear charge pump. They are formed by S1 and C1, S2 and C2, S3 and C3. Each pair of (VS, VD) is summarized in Table 1. The largest time constant limits the knee frequency of the entire charge pump and for the 3-stage linear charge pump the worst VS+VD is given by Vbuffer+2Vsource. As a consequence, the critical path composed of S3 and C3 bounds the maximum switching frequency for complete charge sharing.

Table 1. Drain and source voltages before charge sharing

Voltages before the charging sharing Groups VS VD VS+VD

S1 and C1 Vsource V buffer-3Vsource V buffer-2Vsource

S2 and C2 2Vsource V buffer-2Vsource V buffer

S3 and C3 3Vsource V buffer-Vsource V buffer+2Vsource

B. Proposed Tree Topology Charge Pump A tree topology 4× charge pump is proposed to improve

the charge transfer capability which is shown in Figure 7. The front-end stages (CP1 and CP2) operate like voltage doublers and the back-end stage (CP3) makes use of the already boosted voltages to output an even higher voltage. Compared to the 4× linear topology in Figure 5, this structure has the same hardware cost, i.e., 12 MOS transistors and 3 capacitors.

Figure 7. Proposed tree topology 4× charge pump

First, let us discuss the operation mechanism of the proposed charge pump. Assume Φ is high in the first half clock cycle, thus, the bottom plate of C3 is discharged to zero due to closed S10. The top plate of C3 is charged up by CP1 through S2. At the end of this half cycle, voltages at nodes N1 and N5 are stable at the same value, if complete charge sharing occurs. Let us denote this voltage value as VX1 and hence the amount of charge stored in CP3 is C3VX1. In the next half cycle, when Φ is low and S2 is open, CP1 and CP3 are isolated. But node N4 will be charged up by CP2 through S6. This voltage denoted as VX2 is stable in the end. Thus, the potential of node N5 becomes VX1+VX2. If it is higher than Vbuffer, the charge stored in C3 within the previous half cycle will be transferred to the energy buffer through S9. The charge sharing through S9 ceases when the potential of node N5 decreases to Vbuffer. In the next paragraph we develop the expression for steady state output current in the proposed charge pump.

Assuming that the switching frequency is low enough, there is complete charge sharing within the applied clock cycle. Let each capacitor value be denoted by C. During each clock cycle, the amount of charge transferred to the energy buffer by the voltage doubler CP3is given by:

The amount of charge transferred from CP1 to CP3 is:

The amount of charge transferred from CP2 to CP3 is:

According to the analysis results in [10] [11], for the one-

stage charge pump CP3, when Q3 amount of charge is being transferred to the output energy buffer, CP1 has to provide Q3 amount of charge by direct charging and CP2 has to provide Q3 amount of charge to the bottom plate of C3. Hence, Q1=Q2=Q3, from which we obtain the following:

The output current flowing into the energy buffer is:

In contrast with (1), the derived output current equation for tree topology charge pump is the same as that for linear charge pump. Hence, we can infer that both topologies have the same charge transfer capability when the switching frequency is low and complete charge sharing is guaranteed. Inspecting the charge sharing paths in Figure 7, we obtain each pair of (VS, VD) as shown in Table 2. The largest time constant exists in the path of S2 and C3, which limits the complete charge sharing at higher switching frequencies in the tree topology.

Compared with Table 1, the tree topology has a relatively smaller VS+VD, which results in a lower on-resistance of the charge sharing path and an intrinsic higher knee frequency. Figure 8 illustrates how the charge pump performance varies with the switching frequency. It can be seen that the output currents coincide with each other at lower frequencies while the tree topology has a higher knee frequency.

Equations (1) and (7) also point out the minimum input voltage required for the normal operation (i.e., positive output current) of 4× charge pumps is a quarter of the energy buffer voltage. When the input voltage is equal to or lower than 0.25Vbuffer, the charge pump is unable to harvest charge into the energy buffer, which reflects its loss of charge transfer capability.

Table 2. Drain and source voltages before charge sharing

Voltages before the charging sharing Groups

VS VD VS+VD

S1 and C1 V source V buffer-3V source V buffer-2V source

S5 and C2 Vsource V buffer-3Vsource V buffer-2Vsource

S2 and C3 2Vsource V buffer-2Vsource Vbuffer

Figure 8. Knee frequency comparison between the two topologies

C. System Level Power Harvesting Performance In this section, we discuss the power harvesting efficiency

from a system perspective. As shown in Figure 4, the charge pump operates with the help of the control unit, which also consumes power from the energy buffer. In order to calculate the effective harvested power, the power loss due to the control unit should be taken into account and deducted from the charge pump output power. Dynamic power consumption

dominates the total power loss of the control unit and is modeled as:

The average current loss due to the control unit is:

The current loss linearly increases with the switching frequency. The overall power harvesting performance is plotted in Figure 9, where the solid and dashed line represent the charge pump output current and control unit current loss, respectively. When the switching frequency is low, the charge pump output current overwhelms the control unit current loss, so the effective harvested output current increases linearly. When the applied switching frequency is over the knee frequency, the charge pump output current gradually saturates due to incomplete charge sharing. Since the control unit current loss increases linearly with switching frequency, hence, the effective harvested output current will reach its maximum system power transfer point (MSPTP), as shown in Figure 9. Application of excessive switching frequency results in the control unit current loss exceeding the charge pump’s output current negating the effect of harvesting.

Figure 9. Illustration of the net harvested power

IV. SIMULATION RESULTS

The proposed tree topology charge pump was implemented in IBM 65nm CMOS technology and its layout is shown in Figure 10. Its dimensions are 680µm×400µm. HSPICE simulations were carried out using the MOSFET BSIM models. The transistors in both charge pump topologies were designed with the same W/L ratios of 200µm/60nm. A rechargeable battery was used as the energy buffer and its voltage was fixed at 1V. On-chip MOS gate capacitors (500pF) were used to realize a compact area. As discussed in the previous section, in order to achieve energy harvesting by a 4×-charge pump, the input voltage cannot be lower than one fourth of the energy buffer voltage. As a consequence, three input voltages (0.28V, 0.3V, and 0.33V) were used for simulations. The transistor threshold voltage is around 0.43V, which indicates that these input voltages are sub-threshold.

Figure 10. Layout of the proposed power converter design

Our first simulation experiment verified the relationship between the charge pump switching frequency and its output current. The switching frequency was swept from 1.66MHz to 100MHz. The simulated charge pump output currents are plotted in Figure 11. When the switching frequency is in the range 1.66MHz - 3.33MHz, both charge pumps can guarantee complete charge sharing within the clock cycle. We can observe that the output current is proportional to the applied switching frequency and both output currents have almost similar values. However, when the switching frequency increases further, the advantage of the tree topology becomes evident. The output current of tree structure is much higher than that of the linear structure in the range of 3.33MHz - 100MHz. Thus, these simulation results validate our analysis and prediction given in Section III. Figure 12 shows the percentage improvement in terms of the charge pump output current. When the frequency is higher than 20MHz, our proposed tree topology has around 20% improvement.

Figure 11. Charge pump output current vs. switching frequency

Figure 12. Percentage improvement in output current

The second simulation evaluated the system power harvesting performance. Figure 13 shows the simulated control unit current loss. Taking into account the power overhead due to control unit current loss, we obtained the simulation results for the effective harvested power for two charge pump structures plotted in Figure 14. For example, when the input voltage is fixed at 0.28V, the effective

harvested current increases with increasing switching frequency, and reaches its peak value at around 20MHz. Beyond this point, it drops slowly due to the current overhead of extensive switching. Table 3 summarizes the maximum effective output currents and improvement percentages for three input voltage levels. On average, the proposed design has a 20-30% improvement over the linear topology. The results in Table 3 also indicate that the improvement due to tree topology increases as input voltage becomes smaller.

Figure 13. Current loss in the control unit

Figure 14. Effective harvested current into the energy buffer

Table 3. Maximum effective output current for the Linear and Tree topologies and the percentage improvement

Vsource Maximum effective

output current (Linear) Maximum effective

output current (Tree) Percentage Improvement

0.28V 146.05µA 190.47µA 30.41%

0.3V 295.39µA 378.79µA 28.23%

0.33V 550.79µA 681.12µA 23.66%

V. CONCLUSION

Energy harvesting has emerged as a feasible and attractive option to improve battery lifetime in micro-scale electronic systems such as biomedical implants and wireless sensor

nodes. This paper presented the design of an optimized power converter for ultra-low voltage energy transducers. The design is based on a tree topology charge pump that has a reduced charge sharing time, leading to an improved charge transfer capability. The design has been implemented using 65nm technology and simulations demonstrate that the design results in an increase of up to 30% in harvested power compared to existing linear charge pump designs.

REFERENCES [1] B. Atwood, B. Warneke, and K. S. J. Pister, “Preliminary circuits for

smart dust”, Southwest Symposium on Mixed Signal Design, pp. 87-92, 2000.

[2] R. F. Yazicioglu, et al., “Ultra-low-power biopotential interfaces and their applications in wearable and implantable systems”, Microelectronics Journal, vol. 40, no. 9, pp. 1313-2321, 2009.

[3] L. Mateu and F. Moll, “Review of energy harvesting techniques and applications for microelectronics”, SPIE Microtechnologies for the New Millennium, pp. 359-373, 2005.

[4] V. Raghunathan and P. H. Chou, "Design and Power Management of Energy Harvesting Embedded Systems", ACM/IEEE International Symposium on Low Power Electronics and Design (ISLPED), pp. 369-374, Oct. 2006.

[5] L. Mateu, et al., “Energy harvesting for wireless communication systems using thermogenerators,” Conference on Design of Circuits and Integrated Systems (DCIS), 2006.

[6] K. L. Chu, et al., “A nanoporous silicon membrane electrode assembly for on-chip micro fuel cell applications”, Journal of Microelectromechanical Systems, vol.15, iss. 3, pp. 671-677, 2006.

[7] J. W. Kimball, T. L. Flowers and P. L. Chapman, “Issues with low-input-voltage boost converter design”, Power Electronics Specialists Conference, vol.3, pp. 2152- 2156, 2004.

[8] Dickson J. F., “On-Chip high voltage generation in MNOS integrated circuits using an improved voltage multiplier technique”, IEEE J. Solid State Circuits, vol. SC-11, no. 3, pp. 374-378, June 1976.

[9] F. Su, W. H. Ki and C. Y. Tsui, “Gate control strateies for high efficiency charge pumps”, IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1907-1910, 2005.

[10] F. Pan and T. Samaddar, Charge Pump Circuit Design, McGraw-Hill Professional, 2006.

[11] H. Shao, C-Y. Tsui and W. H. Ki, “A micro power management system and maximum output power control for solar energy harvesting applications", ACM/IEEE International Symposium on Low Power Electronics and Design (ISLPED), pp.298-303, 2007.

[12] P.Favrat, P. Deval, and M. J. Declercq, “A high-efficiency CMOS voltage doubler”, IEEE J. Solid State Circuits, vol. 33, no. 3, pp.410-416, 1998.

[13] T. Ying, W. H. Ki, and M. Chan, “Area-efficient CMOS charge pumps for LCD drivers”, IEEE J. Solid State Circuits, vol. 38, no. 10, pp. 1721-1725, 2003.

[14] C.Y. Tsui, H. Shao, W.H. Ki, and F. Su, "Ultra-low voltage power management circuit and computation methodology for energy harvesting applications", ACM/IEEE Asia and South Pacific Design Automation Conference (ASP-DAC), pp. 96-97, 2006.

[15] D. Brunelli, L. Benini, C. Moser, and L. Thiele, “An efficient solar energy harvester for wireless sensor nodes”, ACM/IEEE Design, Automation, and Test in Europe (DATE), pp. 104-109, 2008.

[16] Solar world Inc., http://www.solar-world.com/ [17] Micropelt Inc., http://www.micropelt.com/ [18] Mypelt simulation tool., www.micropelt.com/products/mypelt.php