Analysis of a Multi-Ratio Switched Capacitor DC-DC Converter for a Supercapacitor Power Supply

. An energy harvesting system can use a supercapacitor in order to store energy; however, a voltage regulator is required to obtain a constant output voltage as the supercapacitor discharges. A Switched-Capacitor DC-DC converter allows for complete integration in CMOS technology, but requires several topologies in order to obtain a high efficiency. This paper presents the complete analysis of these topologies in order to determine expressions that allow to design and determine the optimum input voltage ranges for each topology. These expressions are verified using electrical simulations.


Introduction
In order to achieve infinite operation, electronic systems must obtain their energy directly from the surrounding environment [1].This kind of procedure is commonly known as energy harvesting.Depending on the environment where the system is located at there are different energy sources that can be harvested [2].In the case of systems with small size, the available energy is necessarily reduced, typically only providing a fraction of the necessary power needed for the operation of the system.Therefore, such a system has to be powered down while it harvests enough energy in order to work during a short time and then this cycle is repeated.This type of operation is useful for wireless sensor nodes, where the node only needs to report data periodically.This operation requires an energy storing device, which can be either a rechargeable battery or a supercapacitor.A rechargeable battery has the advantage of having a larger energy storing capacity, but its lifetime is seriously reduced by the number of charge/discharge cycles (typically the maximum number is around 1000 cycles).A supercapacitor has the advantage of having limitless charge/discharge cycles and being less expensive than a battery, but it stores much less energy than a battery for the same volume [3].Depending on the mode of operation of the energy harvesting system, a battery, a supercapacitor or both can be used as the energy storing device.In the case of low power remote sensor that uses a power-down power-up cycle, it makes more sense to use a supercapacitor as the energy storage.
The energy stored in a capacitor is given by = 1 2 ⁄ • • , this means that as the capacitor supplies energy to the circuit its output voltage drops.As an example a 1 F supercapacitor charged to its maximum voltage (usually 2.3 V) can store 2.645 J, if this capacitor supplies energy for a circuit with a supply current of 20 mA during 10s, its output voltage would drop ∆ = • = 200 mV.Since the power supply voltage of the circuit should be constant it is necessary to have a voltage regulator between the supercapacitor and the circuit.In order to maximize the energy retrieved from the supercapacitor it is necessary for the voltage regulator circuit be able to both downconvert and up-convert the input voltage with an efficiency as high as possible.This prevents the use of a linear voltage converter because it can only step-down the voltage and its efficiency is reduced when the input voltage is much larger than the output voltage.The voltage regulator can be implemented using either an inductor based or a capacitor based DC-DC converter.The first option requires an inductor which is not feasible to implement in a CMOS integrated circuit.The second option can be fully integrated in a CMOS integrated circuit; however, switched-capacitor (SC) based DC-DC converters have maximum efficiency for a specific input output voltage ratio.This means that in order to maintain a high efficiency the SC voltage converter circuit has to change its topology as the input supercapacitor voltages decreases [4].Therefore a careful analysis of each converter topology should be carried out in order to determine its efficiency as a function of the input voltage and the different parasitic capacitance of the circuit.This analysis will determine the voltages values that will result in a change in the voltage conversion factor.

Relationship to Cloud-based Solutions
Cloud-based solutions rely on several servers, which are connected together through specialized connections, to distribute data processing tasks among them to overcome limitations in environments where the processing and information storage capability is lacking.
Wireless sensor networks which are integrated into cloud services, e.g.wireless sensor networks for real-time data collection, usually rely on batteries to supply energy to the nodes of the network.Due to the limited lifecycle of batteries, the long term sustainability of the wireless sensor network is compromised.An alternative is to use supercapacitors that acquire energy through energy harvesting systems.Unlike a battery, supercapacitors can charge/discharge an unlimited amount of times, however, a voltage regulator is needed to obtain a constant output voltage as the supercapacitor discharges.This paper analyses several SC DC-DC converters topologies in order to determine the optimum input voltage ranges for each topology.

Analysis of the SC DC-DC Converter Topologies
As previously mentioned, a SC based DC-DC converter has maximum efficiency for a given conversion factor between its input and output voltage.This means that as the supercapacitor discharges and its output voltage drops while the output voltage remains constant, it is necessary to use different topologies for the SC converter circuit in order to maintain high efficiency.There is a compromise between the number of different topologies and the complexity of the circuit.In this case the input voltage can vary from 2.3V to 0 V and the output voltage should be held constant around 1 V. Assuming that a supercapacitor with 1F is used, the available energy, for the maximum voltage, will be 2.645 J.When the capacitor voltage is equal to 1 V the energy remaining in the capacitor is still 0.5 J and when it is equal to 0.7 V the remaining energy is only 0.245 J.This means that there is a small payoff in trying to up-convert input voltages smaller than 0.7 V, which would require a circuit with a voltage conversion factor larger than 2. After analyzing the efficiency of several voltage conversion circuits for the different input voltages, it was decided to use only 4 SC capacitor voltage converter circuits.These are depicted in Fig. 1 and correspond to the voltage conversion factors of 1/2, 2/3, 1 and 3/2 [4].Each of these converters is responsible for a given range of the input voltage (where its efficiency is maximized) and its clock frequency is adjusted in order to obtain an output voltage of 1 V at its output, independently of the input voltage and of the load.These 4 topologies can be combined into a single circuit with 3 capacitors that can be configured to any of the 4 topology using switches.In order to determine what are the input voltage ranges that maximize the efficiency for the different topologies it is necessary to analyze the behavior of each SC converter circuit to determine its operating parameters (efficiency and output voltage) as function of the different design parameters (capacitance values, clock frequency, etc) and the different non-ideal effects (e.g.parasitic capacitance, switch ON resistance).

Analysis of the Efficiency of the 1/2 SC Converter
Assuming that the clock frequency is adjusted in order to have the desired output voltage (e.g.= 1 V) and that output decoupling capacitor is large enough, it is reasonable to assume that the output voltage value is constant.In this case, the schematics of the 1/2 SC converter circuit during phase and are shown in Fig. 2 and Fig. 3 respectively.In these schematics, and represents the bottom and top plate parasitic capacitances of and represents the gate capacitance of all the CMOS switches in the circuit.In each clock cycle, the switches drain a charge (∆ ) from through the clock buffers.Using conventional switched-capacitor circuit analysis techniques [5] (Chapter 5) it is possible to determine the charge in each capacitor at the end of each clock phase.The resulting equations are shown in (1) and these allow to determine Δ , Δ , Δ , and Δ .
The input (' ) and output (' ) power are calculated using Δ , Δ , Δ , and Δ and the values of and and are given by The efficiency (-) is defined as the ratio between ' and ' ,with = 1 V , it is given by.
The 1/2 SC converter circuit was simulated in Spectre with ideal switches and capacitors for different input voltages between 2 to 2.5 V and with fixed at 1 V. From these transient simulations its efficiency was calculated and compared to the efficiency calculated using (3)

Analysis of the Output Voltage of the 1/2 SC Converter
A similar analysis can be performed in order to determine the output voltage of the SC converter, assuming that its output is connected to a load resistor (1 ) The resulting equations are shown in (4).Solving in order to 607 and then considering the steady state condition ( 607 = 60 − 17 = ), allows to calculate (5).In order to simplify (5) it is possible to assume that ≫ resulting in (6).Fig. 5 represents as a function of the frequency for = 1 : the solid lines are the plot of ( 6) and the marks correspond to the simulation results.These validate expression (6).This graph shows that it is possible to control the value of by adjusting the clock frequency.

Circuit Design
In the previous section a theoretical study of the 1/2 converter was carried out.The behavior of the other three converters was determined using the same methodology.The 1/1 converter, is only affected by the top plate parasitic capacitance ( ), because the bottom plate capacitance is shorted to ground in both phases.Since is charged and discharged in parallel with , there is no charge lost to ground through .This means that the efficiency of this converter is not degraded by .Only the gate capacitance from the switches causes a decrease in the efficiency.
The 2/3 and 3/2 converters have a similar topology except for the switches that connect to and to .In these converters, there are five parasitic capacitances: , , , and Q .Where the number 1, 2 and 3 refers to , and Q , respectively.The analysis showed that Q has no impact on the efficiency.Similarly to the 1/2 converter, this is because Q charges and discharges in parallel with Q in both phases.The other four capacitances result in a decrease in -.
In order to better understand the design constraints of the circuit, the circuit will be designed using a 130nm CMOS technology and the capacitors are implemented using MIMCAP including parasitics.Limiting the total area for the capacitors to 1 mm 2 results in a maximum capacitance value of 1 nF this means that 1/2 and 1/1 converters will have a capacitor of 1 nF; and the 2/3 and 3/2 converters will have 3 capacitors with 0.33 nF.This type of capacitors have a top and bottom plate parasitic capacitance of, approximately, 0.4% and 0.6% of the nominal value, respectively.
Equivalent equations to (6) were obtained for the remaining 3 converters using the same methodology.These equations are used to calculate for the 4 converters for ' = 10 mV (1 = 100 Ω), is 2, 3/2, 1 and 2/3, according to the converter ratio.
is plotted in Fig. 7 using the parasitic capacitances values of the selected technology.This plot shows that the 2/3 and 3/2 converters need a higher clock frequency to have = 1 , these higher frequencies will lead to lower efficiency due to the power needed to drive the larger switches.
The analysis described in section 3.1 was applied to the remaining 3 converters and their efficiency was calculated.Fig. 8 shows the plot of R in function of G ST .The solid lines are obtained from the equations and the shapes are simulation results.These plots were obtain for U V = W.The efficiency graph of Fig. 8 was calculated independently of the clock frequency and ignoring the charge lost through the gate capacitance of the switches.Since the gate capacitance increases with the CMOS switch size, which increases with the clock frequency value, the efficiency also depends on the clock frequency.The clock frequency value for each converter as a function of the G HIJ in order to have G HIJ = G is shown in Fig. 9.This graph was obtained using similar expressions to (7) calculated for all the converters.The CMOS switches are constituted by NMOS and PMOS devices in parallel.These transistors are sized in order to have an ON resistance (1 ef ) value that results in a time constant compatible with the clock frequency value.In [6] it is shown that 1 ef and for a settling error !ghhih) and a frequency clock !) *+, ) and the capacitance coupled to the switch ! ) are given by: Where n o = 806.24Ω ⋅ μm and n * = 7.1 fF/μm are coefficients derived by simulation Spectre [6].Replacing (9) in (3) resultsas function of , , ghhih, 1 and ) *+, .This now takes into account the effect of the frequency in the parasitic capacitances of the switches.Replacing (7) in this new expression results in in function of , , ghhih, 1 .This final equation describes the efficiency with the frequency adjusting to the value necessary to maintain = 1 V and the extra power consumed by the switches due to this adjustment.Fig. 10 shows the plot ofas function of with ' = 10 mW !1 = 100 Ω), = 1 V, settling error of 1%, = 0.6% and = 0.3% , for the 1/1 and 1/2 = 1 nF; and for the 2/3 and 3/2 = 0.33 nF.The plot shows the maximum achievable efficiency and the corresponding value of .Through this it is possible to determine the range of operation for each converter.For example, in this case the 1/2 converter works from 2.3 to 2.07 V, with a maximum of 84% and a minimum of 64% efficiency.The 2/3 converter works from 2.07 to 1.76 V, with a maximum of 65% and a minimum efficiency of 57%.The 1/1 converter works from 1.76 to 1.05V, with a maximum of 83% and a minimum efficiency of 57%.Finally, the 3/2 converter works from 1.04 to 0.67 V, with a maximum efficiency of 84%.

Conclusion
This paper presented a theoretical analysis of a multi ratio SC DC-DC converter for a supercapacitor power supply of an energy harvesting system.The different SC circuits of the converter were analyzed in order to obtain expressions that allow computing their efficiency as a function of the input voltage.These expressions were verified using electrical simulations and then used to design the circuit for a 130 nm CMOS technology and determine the optimum input voltage ranges for each topology.

Fig. 1 .
Fig. 1.Simplified schematics of the selected topologies for the SC converters

Fig. 4 .
Fig. 4.These simulation results prove that (3) accurately describes the behavior of the efficiency of the converter and validates the theoretical analysis.

Fig. 4 .
Fig. 4. Efficiency of the 1/2 SC converter circuit as a function of and of = = .

Fig. 8 .
Fig. 8. d as function of K LM for the four converters with X Y = W. Z% X and X \ = W. ]% X