ON THE DESIGN OF SWITCHED-CAPACITOR CONVERTERS FOR LOW

POWER APPLICATIONS

By

Yaqub Alhussain Mahnashi

A DISSERTATION

Submitted to

Michigan State University

in partial fulfillment of the requirements

for the degree of

Electrical Engineering — Doctor of Philosophy

2018

ON THE DESIGN OF SWITCHED-CAPACITOR CONVERTERS FOR LOW

ABSTRACT

POWER APPLICATIONS

By

Yaqub Alhussain Mahnashi

Power conversion is crucial to most electronic systems due to different voltage levels re-

quirement within the system. Switched-capacitor converters are a subset of switching mode

converters, where the energy is transferred from the input to the output using only capaci-

tors and switches. SC converters provide compact solution due to its magnetic-less structure,

which makes them ideal to fulfill the high density requirement in low power applications. In

literature, Dickson SC converter is considered a good candidate for low power applications

compared to other SC topologies. Yet, Fibonacci SC (FSC) topology can be competitive

due to their minimum area requirement and relatively good efficiency performance. In ad-

dition, FSC provides the highest voltage-conversion-ratio (VCR) using the least number of

components, and hence can be useful in high gain applications. Therefore, the main goal of

this dissertation is to introduce FSC as a viable solution for low power applications. This

dissertation contributes in three domains: (1) SC theory, (2) FSC synthesis and (3) FSC

design and implementation.

For the first part, the well-known SC fundamental limit theory, proposed in 1995 by Mark

S. Makowski et al, is revisited. This theory sets the VCR boundaries for a specific number

of k flying capacitors in a SC converter. Although this limit is valid for the positive VCR,

we found that another condition must be satisfied for the negative VCR. As a result, we

propose a generalized version of the theory to overcome these limitations, which establishes

the foundation for the rest of the work in this dissertation. Second part of the dissertation

explores the FSC synthesis problem. In fact, synthesizing nonlinear SC converters like FSC

to achieve certain/multiple VCR is not trivial, and usually performed using ad-hoc approach.

Therefore, an efficient algorithm to address the SC synthesis problem is proposed, which is

made available as an open-source tool, called FSC Synthesizer, to promote the use of FSC.

The proposed tool is verified by implementing FSC converter that achieves four VCRs.

Finally, the design procedure and analysis of a variable FSC converter is investigated

in both, discrete and on-chip implementation. For the first part, discrete FSC design is

studied, and a prototype for a variable FSC is built and tested. The PCB occupies an area of

2.7inch2, which includes the controller and FSC. The adaptive VCR technique is employed to

regulate the output voltage. For 300-600mV input voltage, the proposed converter provides a

mean output voltage of almost 1.46V with a 2.67% discrepancy from 1.5V nominal designed

voltage. For the second part, the on-chip integration challenges including parasitic charge

loss and start-up from low voltage, are studied. A charge recycling technique exploiting the

presence of parasitic capacitors in each other phase of FSC converter has been employed

resulting in 27% reduction in the power loss, which improves the overall efficiency by 12.7%.

For a proof-of-concept, a monolithic FSC is implemented in 0.5µm CMOS technology on

4mm2 die area. The post-layout simulation is carried out using Virtuoso ADE. With 0.3V

input voltage, the system achieves 47.5% peak power efficiency with 5µW load.

Copyright by
YAQUB ALHUSSAIN MAHNASHI
2018

This dissertation is dedicated to the memory of my beloved father, Alhussain Mahnashi,

who always believed in his son. Rest in peace, father. . .

v

ACKNOWLEDGMENTS

I would like to take this opportunity to express my sincere appreciation for all the support and

encouragement that have led to the completion of this dissertation. I am greatly indebted

to my advisor, Prof. Fang Z. Peng, for his continuous support, help and encouragement

throughout my PhD studies at Michigan State University.

I would also like to thank Prof. Bingsen Wang, Prof. Wen Li, and Prof. Chunqi Qian

for serving on my committee. I am deeply grateful to them for their valuable comments and

insightful discussions. I owe a sincere acknowledgment to Saudi Arabia government through

King Fahd University of Petroleum and Minerals to grant me a prestigious scholarship to

peruse my PhD studies.

Special thanks go to my colleagues in the Power Electronics and Motor Drives Lab: Dr.

Shuitao Yang, Dr. Gujing Han, Dr.Wei Qian, Deepak Gunasekaran, Ujjwal Karki, Hulong

Zeng, Xiaorui Wang, Yunting Liu, Allan Taylor, Nomar Goz´alez-Santini, Omar Aljawhari

and Petros Taskas, for making my stay at Michigan State University more enjoyable, and

for their insightful and helpful academic inputs.

I am greatly grateful to my family and friends for their tremendous support and encour-

agement. My profound gratitude is to my mother for her endless love, prayers and support. I

could not be more grateful to my beloved wife and sons, Yusuf and Albara. This dissertation

would not have been accomplished without their love, care, patience and support. Thank

you all for making this dream a reality.

vi

TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

ix

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

x

KEY TO ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

Chapter 1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 A Brief History on SC Development . . . . . . . . . . . . . . . . . . . . . . .
1.3 Research Scope and Contributions . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Report Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 2

SC Converter Fundamental Theory . . . . . . . . . . . . . . . . .
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1
. . . . . . . . . . . . . . . . . . . . . . . . .
2.2 General SC Fundamental Limit
2.3 MIMO SC converter
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 3

SC Converter Synthesis . . . . . . . . . . . . . . . . . . . . . . . .
3.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Proposed FSC Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Case Study and Results
3.3.1 FSC Converter with VCR=30 . . . . . . . . . . . . . . . . . . . . . .
3.3.2 Variable FSC Converter with VCR=5, 4, 3, 5/2 . . . . . . . . . . . .
3.3.3 Variable FSC Converter with VCR=1, 3/4, 2/3 . . . . . . . . . . . .
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.1 Theoretical Analysis

Chapter 4 Reconfigurable FSC Converter: Discrete Design . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.1 Converter Equivalent Resistance . . . . . . . . . . . . . . . . . . . . .
4.1.2 Power Loss Analysis
. . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.3 Output Regulation Methods . . . . . . . . . . . . . . . . . . . . . . .
4.2 Converter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.1 Power Loss Calculation . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.2 Components Selection . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.3 Control Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
1
5
8
8

10
10
12
14
19

20
20
23
27
27
28
29
33

34
34
35
40
43
44
45
48
49
54
55

vii

Chapter 5

5.3

5.2

Integrated FSC Converter . . . . . . . . . . . . . . . . . . . . . . .
5.1 On-chip Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Switch Implementation
. . . . . . . . . . . . . . . . . . . . . . . . .
5.1.1
IC Solar Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.2
IC Capacitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IC FSC challenges
5.2.1
Startup Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.2 Parasitic Capacitor . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IC FSC Converter Analysis
. . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Die Area Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.2 Power Loss Analysis
. . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4
IC FSC Converter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 6 Analog and Digital Circuit Design . . . . . . . . . . . . . . . . .
6.1 Cross-Coupled Charge Pump Design . . . . . . . . . . . . . . . . . . . . . .
6.2 Clock Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Comparator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IO Pad Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.4
6.5 Final Layout and Post-Layout Results
. . . . . . . . . . . . . . . . . . . . .
6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 7 Conclusions and Future Work . . . . . . . . . . . . . . . . . . . .
7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

56
56
57
60
61
63
63
65
67
67
68
69
74

75
76
79
82
88
89
93

97
97
98

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

viii

LIST OF TABLES

Table 3.1:

All possible realizations for FSC with VCR=5/3 . . . . . . . . . . . . .

22

Table 3.2:

Comparison of SC converter synthesis techniques . . . . . . . . . . . . .

23

Table 3.3:

All possible realizations for VCR=30/1 . . . . . . . . . . . . . . . . . .

28

Table 3.4:

Optimized realization for variable FSC with VCR=5, 4, 3, 5/2 . . . . .

30

Table 3.5:

LUT for converter synthesized by proposed method . . . . . . . . . . .

31

Table 3.6:

LUT for converter in [21]

. . . . . . . . . . . . . . . . . . . . . . . . . .

33

Table 4.1:

Variable FSC converter design specifications

. . . . . . . . . . . . . . .

44

Table 4.2:

Different chimerical switches with their specifications.

. . . . . . . . . .

49

Table 4.3:

LUT for variable FSC converter. . . . . . . . . . . . . . . . . . . . . . .

52

Table 5.1:

Ambient sources.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

64

Table 5.2:

C5 technology parameters

. . . . . . . . . . . . . . . . . . . . . . . . .

71

Table 6.1:

Truth table for N-T DFF.

. . . . . . . . . . . . . . . . . . . . . . . . .

86

Table 6.2:

IC pin configurations. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

90

Table 6.3:

Performance Comparison with SC converters in the literature.

. . . . .

95

ix

LIST OF FIGURES

Figure 1.1: Types of DC-DC converters.

. . . . . . . . . . . . . . . . . . . . . . . .

Figure 1.2:

Linear regulator (LDO) circuit.

. . . . . . . . . . . . . . . . . . . . . .

Figure 1.3: Functional SC converter building blocks.

. . . . . . . . . . . . . . . . .

Figure 1.4: Conceptual smart cell building blocks. . . . . . . . . . . . . . . . . . . .

Figure 1.5:

(a) Cockroft-Walton multiplier, (b) Dickson charge pump. . . . . . . . .

2

2

4

5

7

Figure 2.1: One cell FSC converter. . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

Figure 2.2: VCRs for one flying capacitor FSC converter in steady state.

. . . . . .

14

Figure 2.3: General structure for FSC converter.

. . . . . . . . . . . . . . . . . . .

15

Figure 2.4: Circuit realization for FSC converter with dual outputs. . . . . . . . . .

18

Figure 2.5: Output voltage waveforms for dual outputs FSC converter.

. . . . . . .

19

Figure 3.1: Flow chart for the proposed algorithm.

. . . . . . . . . . . . . . . . . .

26

Figure 3.2: Terminal assignments using FSC Synthesizer for variable FSC with
VCR=5, 3, 4, 5/2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

Figure 3.3: Variable FSC with VCR=5, 3, 4, 5/2.

. . . . . . . . . . . . . . . . . . .

30

Figure 3.4: Buck SC converter with VCR=1, 3/4, 2/3: (a) FSC topology using pro-
. . . . . . . . . . . . . .

posed method. (b) Converter proposed in [21].

32

Figure 4.1:

SC converter steady state model. . . . . . . . . . . . . . . . . . . . . . .

35

Figure 4.2:

SC converter equivalent impedance vs switching frequency.

. . . . . . .

36

Figure 4.3: FSC converter with VCR=3. . . . . . . . . . . . . . . . . . . . . . . . .

37

Figure 4.4: FSC converter with VCR=3: (a) state I, Sa=1, Sb=0. (b) state II, Sa=0
Sb=1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38

Figure 4.5: Gate switching intervals.

. . . . . . . . . . . . . . . . . . . . . . . . . .

42

x

Figure 4.6: Variable FSC with VCR=5, 3, 4, 5/2.

. . . . . . . . . . . . . . . . . . .

45

Figure 4.7: All operation states for proposed FSC: (a) VCR=5, (b) VCR=4, (c)
. . . . . . . . . . . . . . . . . . . . . . . .

VCR=3, and (d) VCR=5/2.

Figure 4.8:

Ploss verses switching frequency for different switches. . . . . . . . . . .

46

49

Figure 4.9: Efficiency for different loads.

. . . . . . . . . . . . . . . . . . . . . . . .

50

Figure 4.10: Variable FSC converter design flow chart using discrete components. . .

51

Figure 4.11: Ring oscillator with non-overlapping clocks generator.

. . . . . . . . . .

52

Figure 4.12: The gearbox signals control circuit.

. . . . . . . . . . . . . . . . . . . .

53

Figure 4.13: Variable FSC converter PCB.

. . . . . . . . . . . . . . . . . . . . . . .

53

Figure 4.14: Vout response to Vin for variable FSC converter.

. . . . . . . . . . . . .

54

Figure 5.1: CMOS structure.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57

Figure 5.2: Photo-diodes available in CMOS process. Reproduced from [36].

. . . .

61

Figure 5.3: On-chip PIP capacitor: (a) simple model, (b) 3D layout structure.

. . .

62

Figure 5.4: Dead-time charge recycling (QR): (a) Timing signals, (b) FSC converter.

66

Figure 5.5:

IC FSC design procedure flow chart. . . . . . . . . . . . . . . . . . . . .

70

Figure 5.6: Proposed FSC with parasitic capacitors.

. . . . . . . . . . . . . . . . .

72

Figure 5.7: Power losses shares with and without charge recycling (QR).

. . . . . .

73

Figure 6.1: FSC IC building blocks. . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

Figure 6.2: Possible FSC controller structures: (a) Feed-forward, (b) Hysteresis con-
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

troller.

76

Figure 6.3: Three-stage RO with eight CCCP modules. . . . . . . . . . . . . . . . .

77

Figure 6.4:

Layout of RO with CCCP.

. . . . . . . . . . . . . . . . . . . . . . . . .

78

Figure 6.5: Post-Layout simulation for the startup circuit using 0.5V input voltage.

79

Figure 6.6: Three-stage CSRO with non-overlapping clock generator.

. . . . . . . .

80

xi

Figure 6.7:

Level shifters with QRU.

. . . . . . . . . . . . . . . . . . . . . . . . . .

81

Figure 6.8: Clocks generation layout. . . . . . . . . . . . . . . . . . . . . . . . . . .

81

Figure 6.9: Clocks waveforms (post-layout simulation).

. . . . . . . . . . . . . . . .

82

Figure 6.10: Comparator functional blocks: (a) Static comparator. (b) Dynamic com-
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

parator.

83

Figure 6.11: CMOS comparator structures.

. . . . . . . . . . . . . . . . . . . . . . .

83

Figure 6.12: Proposed CMOS dynamic comparator.

. . . . . . . . . . . . . . . . . .

85

Figure 6.13: CMOS dynamic comparator layout.

. . . . . . . . . . . . . . . . . . . .

86

Figure 6.14: Post-layout simulation for the proposed dynamic comparator.

. . . . .

87

Figure 6.15: The gearbox controller layout.

. . . . . . . . . . . . . . . . . . . . . . .

87

Figure 6.16: ESD for IO pads.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

88

Figure 6.17: Layout of an IO pad with ESD.

. . . . . . . . . . . . . . . . . . . . . .

89

Figure 6.18: The chip layout with pin configuration.

. . . . . . . . . . . . . . . . .

91

Figure 6.19: Efficiency verses Iout for the proposed FSC converter with 300mV input
voltage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 6.20: Efficiency verses Iout for the proposed FSC converter with 300mV input
voltage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

93

94

Figure 6.21: Vout verses Iout for the proposed FSC converter with 300mV input voltage. 94

Figure 6.22: VCR verses Vin for the proposed FSC converter.

. . . . . . . . . . . . .

95

Figure 7.1: MIMO SC conceptual blocks for possible applications:

(left) on-chip
power management unit for microprocessors, and (right) energy harvest-
ing system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

99

xii

KEY TO ABBREVIATIONS

ADE Analog Design Environment

SC Switched-Capacitor

SI Switched-Inductor

FSC Fibonacci Switched-Capacitor

VCR Voltage Conversion Ratio

aVCR adaptive Voltage Conversion Ratio

LDO Linear Regulator

EMI Electromagnetic Interference

IC Integrated Circuit

PCB Printed Circuit Board

WSN Wireless Sensor Network

CW Cockroft-Walton

CP Charge Pump

LQP Linear Charge Pump (Dickson converter)

MIMO Multiple Input Multiple Output

SIMO Single Input Multiple Output

MISO Multiple Input Single Output

LUT Look Up Table

MOS Metal Oxide Semiconductor

CMOS Complementary Metal Oxide Semiconductor

MOM Metal Oxide Metal

MIM Metal Insulator Metal

PIP Poly Insulator Poly

ANTZ All Negative terminals connected To Zero

SPTT Single Pole Triple Throw

xiii

SSL Slow Switching Limit

FSL Fast Switching Limit

VCE Voltage Conversion Efficiency

PCE Power Conversion Efficiency

ZVS Zero Voltage Switching

RO Ring Oscillator

CSRO Current-Starved Ring Oscillator

EH Energy Harvesting

AQP Auxiliary Charge Pump

NCP1 modified Dickson Charge Pump Version 1

NCP2 modified Dickson Charge Pump Version 2

CCCP Cross-Coupled Charge Pump

TP Top Plate

BP Bottom Plate

xiv

Chapter 1

Introduction

In this chapter, the switched-capacitor (SC) converter is introduced. For lower power applica-

tions, many SC configurations can be utilized. However, Fibonacci SC converter (FSC) [1,2]

and Dickson converter [3] are considered to be the best topologies due to their less area

requirement and good efficiency performance. Therefore, an overview of these topologies

will be presented, and followed by the research scope and summary of contributions.

1.1 Overview

The power conversion is crucial in most electronic systems. As shown in Figure 1.1, there

are two ways to implement DC-DC converters: passive and switching-mode implementation.

The DC-DC converters are used to step-down or step-up the input voltage based on the

system requirement. For step-down conversion, passive implementation or so-called linear

regulators (LDO) are commonly used due to less area requirement and easy control. Figure

1.2 illustrates a simple implementation of LDO, which is basically a voltage divider circuit

implemented around an active controlled switch that works as a variable resistor. Despite

its simplicity and minimal active area requirement, LDO suffers from low efficiency and only

can be used for step-down conversion [4]. Note that, the LDO efficiency can be very high

if the output voltage is close to the input voltage, i.e theoretically 100% efficiency can be

achieved when no conversion is performed (Vin=Vout).

1

Figure 1.1: Types of DC-DC converters.

Figure 1.2: Linear regulator (LDO) circuit.

2

DC-DC ConvertersPassive ConvertersSwitching mode ConvertersSwitched-Capacitor (SC)Switched-Inductor (SI)Liner Regulator (LDO)Resonant converters (hybrid)VoutReqVinVoutM1R1R2VinRLRLVrefReqThe switching mode converters are more versatile as they can be implemented to step-

down or step-up the input voltage [5].

In addition, some converters are capable of both

conversions owing to the intelligent control circuit and converter design. Switching mode

converters are implemented using switch networks (SN) and storage elements to transfer

the input energy to the output. Based on the storage element, the switching converters

can be categorized into three main groups: switched-capacitor (SC), switched-inductor (SI)

and resonant converters (hybrid), see Figure 1.1. Examples of SI converters include buck,

boost, SEPIC, ´Cuk, Flyback converters and many others [6]. They are widely used in

commercial products due to their high efficiency and relatively simple design. However, the

main disadvantage of SI converters is the use of an inductor which is a bulky element and

usually associated with poor EMI performance.

On the other hand, the SC converters, sometimes referred to as charge pumps [7], use

capacitors as storage elements, commonly referred to as flying capacitors, which makes them

attractive for monolithic IC integration due to their compact and magnetic-less features,

the functional blocks diagram is depicted in Figure 1.3. Dickson, Fibonacci, ladder, series-

parallel and voltage doubler and their variants are examples of SC converters. Most SC

converters can be driven in open loop scenario using at least two non-overlapping clocks to

avoid shoot-through states. Despite these advantages, SC converters suffer from poor output

regulation and the conversion efficiency drops drastically when the output voltage deviate

from the target voltage due to the input voltage change and/or loading effect. Therefore,

achieving more voltage-conversion-ratios (VCR) minimizes the error between the output and

target voltage which effectively improve the efficiency. Ideally, a the better conversion perfor-

mance can be achieved by increasing the number of VCR. Yet, synthesizing SC converter to

perform multiple VCR is considered a major challenge, commonly referred to as SC synthesis

3

problem.

Figure 1.3: Functional SC converter building blocks.

Generally, solar energy is abundant and can be utilized to prolong the battery life for

applications that require expensive or invasive battery replacement, such as wireless sensor

nodes (WSN) and bio-implementable devices. To drive this work to a practical implemen-

tation, we consider a smart solar cell which consists of an on-chip solar cell, implemented

using photo-diodes, and SC boost converter as shown in Figure 1.4. The concept of smart

solar cell is to replace conventional solar cell with low open circuit voltage, Voc, with a

system containing the solar cell and boost converter to output higher voltage level that can

recharge a battery or power a system. Also, smart solar cell is expected to be autonomous

meaning, the integrated converter starts up and operate without any external biasing. In
addition, the converter needs to smart enough to detect Voc variations, usually 0.3− 0.6V in

miniaturized solar cells, and apply any technique for output regulation. Referring to Figure

1.4, the smart solar cell supplies the power management unit (PMU) with a dc voltage >1V

which is adequate to power up microsystem in some applications. Then, PMU is responsible

4

flying caps.SN.controllerVinVinLoaddriving signalsVoutto power up the system, WSN or implanted device ..etc, from the smart solar cell. In case

if there is no enough power, PMU bypasses the smart solar cell and supply power to the

system using the battery. For this operation, we need to design a SC converter that can

startup from low voltage and configures itself to provide enough supply voltage and all are

done autonomously and without any external component. There are some challenges that

limits a high efficiency in IC SC converters. Therefore, in this work, we will focus only on the

SC converter block. In the following section, a brief history of SC development is presented.

Figure 1.4: Conceptual smart cell building blocks.

1.2 A Brief History on SC Development

SC converters can be traced back to the first high voltage Cockroft-Walton (CW) multiplier

[8]. This multiplier suffers from high stray capacitance and to have efficient multiplication,

pumping capacitors need to be very large which is impractical in monolithic integration. To

overcome this problem, the CW multiplier was modified by connecting pumping capacitors

in parallel to the input instead of being connected in series. This is a well-known Dickson

5

on-chip solar cellSC converterPMUBatterySystem0.3~0.6V>1VSmart solar cellcharge pump (CP) [3], or sometimes referred to as linear charge pump (LQP) because the
VCR is linear and can take any integer from 1...n, where n ≥ 1. Therefore, CW multiplier

and Dickson CP are quite similar in basic structure and operation, but working in two

different power and voltage levels.

The CW multiplier was developed for high voltage generation of 800kV whereas Dickson

CP was intended for low power applications enabling integration in the same substrate.

Figure 1.5 illustrates both circuits, highlighting the similarities between them. Dickson

basically connected the flying capacitors to the clock instead of being connected in series

as in CW circuit. This decreases the stray capacitance at the junctions, x and y, which

improves the total converter performance. Since then, Dickson CP has been the main voltage

multiplier in IC circuits. The output voltage in Dickson CP can be found using the following

equation:

Vout = n(Vin − VD),

(1.1)

where n is the number of stages, and each stage consists of a diode and flying capacitor. VD

is the diode voltage drop which is approximately 500-700mV, and it equals the threshold

voltage (Vth) if the diode is implemented using a MOS transistor. Dickson converter suffers

from the diode voltage drop. Therefore, to achieve higher voltage level, the diode voltage

drop needs to compensated by adding more stages.

Following Dickson invention, many topologies have been proposed in three main trends:

solving the voltage regulation, decreasing the component stress and shrinking the converter

size. The authors in [1] proposed a SC converter where the VCR follows Fibonacci series,

which since then has been considered the first brick in FSC development. Later, the the-

6

Figure 1.5: (a) Cockroft-Walton multiplier, (b) Dickson charge pump.

oretical limits of SC converter is proved in [2], resulting in two important theorems which

help to set guidelines for developing SC converters. This is well-known as the SC fundamen-

tal limit theory. Also, FSC has been proved to have the least number of number of flying

capacitors and switches to implement specific VCR, which lends itself to high power density

applications. Multi-phase FSC can be also implemented to achieve higher and/or balanced

VCR [9, 10], and the performance limits of multi-phase can be found in [11]. However,

multi-phase requires complicated control unit compared to a simple two on-off states control

in the two-phase FSC converter. Therefore, two-phase FSC converter will be considered

throughout this work.

7

SbSaVDCVout(a)xyCa2Ca1Cb1Cb2SbSaVDCVout(b)Ca2Ca1Cb1Cb21.3 Research Scope and Contributions

The aim of this work is to present theoretical analysis of the SC converter, investigate the

multi VCR SC synthesis, and discuss the design of a variable FSC converter in discrete and

IC implementations targeting low power applications. The contributions of this dissertation

can be summarized in the following:

1. Generalizing the SC fundamental limit theory.

2. Proposing a novel algorithm to synthesize variable FSC converters. To the best knowl-

edge of the author, this is the first generic method to configure a variable FSC converter.

An open-source tool is built based on this algorithm to promote the FSC use.

3. Proposing design procedures and optimization to implement FSC converter in low

power applications for both discrete and IC implementation.

4. Proposing an IC controller with charge recycling technique to minimize the parasitic

charge loss in IC FSC converter. Ideally, this technique decreases the parasitic capaci-

tors power loss by 50%. The concept of charge recycling is not new but implementing

this technique to FSC without transforming the FSC to interleaved structure is pro-

posed here.

1.4 Report Organization

The rest of the report is structured as follows.

In Chapter 2, we present the proposed

generalized equation to the SC fundamental limit. Synthesizing FSC converters is discussed

in Chapter 3 and the proposed algorithm is presented. In Chapter 4 and 5, we present the

8

design and implementation of discrete and integrated FSC converter, respectively. Finally,

the report is concluded and future work is provided in Chapter 7.

9

Chapter 2

SC Converter Fundamental Theory

In this chapter, the fundamental limit theory in SC converter is introduced and studied.

This theory is insufficient to determine the accurate boundaries for all negative VCRs in a

SC converter. This work extends the theory to all possible cases of SC converters resulting

in a general form1. The proposed theory in this chapter will be used later in synthesizing

FSC converter.

2.1

Introduction

The fundamental limit in SC converter design which was proved in [2] is a well-known

theory in the literature. The limit creates the foundation for SC converters design and

synthesis. The fundamental limit theory consists of two theorems that define the voltage-

conversion-ratio (V CR = Vout/Vin) boundaries for k flying capacitors in a SC converter, and

the minimum number of switches to realize that converter. The first theorem is the main

concern in this work. It states that for k flying (pumping) capacitors in an SC converter,

the numerator and denominator of achievable VCR follow the Fibonacci series. They are

bounded by 1 as the minimum limit and Fk+2 as a maximum limit, as shown in the following

1The work presented in the chapter has been published in IEEE Transaction of Power Electronics [12].

10

equation:

Vout
Vin

=

P
Q

,

(2.1)

where 1 ≤ (P, Q) ≤ Fk+2, k is the number of flying capacitors in the circuit and Fk is
the kth Fibonacci number found in the Fibonacci series, F = (1, 2, 3, 5, 8, 13, 21...Fk). The

SC converter that has VCR follows or implement this limit is called FSC converter.

This fundamental limit is a powerful basis to minimize the effort of designing and imple-

menting new SC converter topologies. It also helps providing a figure of merit to compare

between different topologies in terms of achievable VCR and element counts. However, this

limit only discusses the positive VCR boundaries and it does not explore the negative VCR

boundaries which is considered as a first limitation.

In this paper, we propose a general

expression that overcomes this limitation and sets the boundaries for all possible VCRs.

The second limitation of the fundamental limit arises from the focus on single-input-

single-output structure only. However, the multiple-input-multiple-output (MIMO) SC con-

verter structures is desired in some applications. Extending the fundamental limit in (2.1)

to include MIMO concept leads to provide a truly generic theory of the fundamental limit.

This chapter is intended to explore and study these two limitations and propose generic

expressions in both cases using FSC topology because it is the best to represent the theory.

Examples of different cases and the practical implications are also discussed to verify the

theory.

11

2.2 General SC Fundamental Limit

In some applications, negative output voltage polarities are desired. In this case, however,

the fundamental limit fails to define the real boundaries that SC converter can achieve for

the negative VCR. Generalizing the fundamental limit to cover all possible combinations

of VCR is needed. The following expression divided the VCR in two regions, positive and

negative. In both, the VCR still follow the Fibonacci series but the boundaries are different.

As (2.2) describes,

Vout
Vin

=

+ P

− P

Q

Q

1 ≤ max|Q, P| ≤ Fk+2
1 ≤ max|Q, P| < Fk+2

,

(2.2)

negative VCRs are less than positive VCRs for the same converter. Note that, P/Q should

be represented in their simplest form, for example 4/4 is considered as 1/1. The conditions

in (2.2) are used to estimate the absolute minimum number of flying capacitors for desired

VCR. More details about the practical implication of this expression are presented below.

To better understand the expressions presented in (2.2), let us illustrate by using an

FSC converter with one flying capacitor, shown in Fig. 2.1. The flying capacitor can be
connected to Vg, Vo or ground through the terminals t1 − t3. This converter can achieve

(1/2, 1, 2) VCRs neglecting the sign, which can be found using (2.1). The negative VCRs
were left out as a default, assuming that this converter is capable to achieve (−1/2,−1,−2)

voltage conversion, which is incorrect. In fact, (2.2) reveals that with one flying capacitor,
the converter can achieve only one negative VCR, which is −1. The look-up-table (LUT)

for all possible terminal connections is presented in Figure 2.1, showing that this converter
can achieve four distinct VCRs in total, (−1, 1/2, 1, 2).

12

Figure 2.1: One cell FSC converter.

On the other hand, the minimum number of flying capacitors needed to achieve other

negative VCRs (−2,−1/2) can be found using (2.2).
In this case, it can be found that
a Fibonacci number of two or greater is required because max|P, Q|=2. Hence, F4 = 3

is the right limit for this converter, meaning two flying capacitors are needed to achieve
the missing VCRs. Yet another example, if a (−1/4) Fibonacci SC converter needs to be

designed. Applying the above steps resulted in F5 = 5, which means the converter needs at

least three flying capacitors. The design procedure has two steps:

1. Defining the real boundary limit using (2.2), i.e Fk+2.

2. Finding the number of required flying capacitors, which is k.

Spice simulation was conducted using ideal switches with Rdson=10mΩ, fs=500kHz and

15µF flying capacitor. The MOSFET based switches are turned on and off, using 50%

duty-cycle complementary non-overlapping square signals to avoid shoot-through states. All

possible VCRs were plotted in steady-state and depicted in Figure 2.2. The simulation result

agrees with the proposed theoretical expression. One note here, in case if the summation of

P and Q is not a Fibonacci number, the next Fibonacci number is selected. For instance,

13

CSaSbSaSbVoutt1t2t3Vinusing V CR = −1/3, the summation of absolute values of the numerator and denominator is

four which is not a Fibonacci number. Thus, F5 = 5 is assigned as a limit for this converter

instead of four.

Figure 2.2: VCRs for one flying capacitor FSC converter in steady state.

2.3 MIMO SC converter

The MIMO SC converter and its special cases, the single-input-multiple-outputs (SIMO) and

the multiple-input-single-output (MISO), are discussed here. As mentioned earlier, some

applications require multiple outputs generated from one or multiple inputs, like the power

management in microprocessors [13]. Usually, different DC-DC converters are employed

to generate these different output levels. On the other hand, SIMO SC converters can

do the job effectively achieving a real compact solution. Considering MISO SC converter

implementation, a circuit for low power energy harvesting is used as an example. In this

circuit, a single point-of-load is usually powered by multiple ambient sources to harvest

more power. It can either be implemented by using different conventional SC converters,

14

0.20.40.60.80.1Time (ms)-1012VCRor by simply using the MISO SC converter which offers better solution in terms of power

density. However, synthesizing these converters without knowing the VCR boundary limits

is a tedious job. MIMO implementation is proposed in this paper by employing the terminal

weight concept found in [14]. Although MIMO implementation can be extended to all

SC converter topologies, only the MIMO Fibonacci SC converter is considered because the

interest here is to investigate the fundamental limit theory.

Figure 2.3: General structure for FSC converter.

The general structure of the MIMO FSC converter is depicted in Figure 2.3. It consists

of two main blocks: the converter core, which is composed of FSC cells, and a pool of voltage

nodes. The voltage nodes are connected to the core using terminals. The number of these

terminals depend on the number of available FSC cells in the core. For instance, the number

of terminals in k cells FSC converter equals k + 2. These terminals are assigned a weight

that depends on its position in the circuit. There are four conditions need to be met to

15

C1t1t2SaSbSaC2t3SbSaSbCktk+1Sa,bSb,aSa,btk+2Sb,agearbox voltage nodes (Vin, Vout, gnd)C1t1t2SaSbSaC2t3SbSaSbCktk+1Sa,bSb,aSa,btk+2Sb,agearbox voltage nodes (Vin, Vout, gnd)Basic FSC cellassign the terminal weight and achieve FSC converter structure [14]. These conditions are

as follows:

1. w1 = Fk+2,

2. wk+2=−1,

3. (cid:80)k+2

i=1 wj = 0,

4. |wj| = (1, 1, 2 . . . Fk+2).

The first condition implies that the first terminal (t1) has a positive weight equals to

the maximum achievable VCR, which is Fk+2. The other terminals are assigned negative

weights using the b-d conditions. Note that all terminal weights follow the Fibonacci series

as suggested in step (4).

Suppose that the MIMO FSC converter consists of n inputs and m outputs. With careful

terminal weight assignment, the following equation relates the input and output voltages,
where n+m ≤ k + 2:

n(cid:88)

i=1

wvgi

vgi +

m(cid:88)

j=1

wvoj

voj = 0.

(2.3)

This equation deals with MIMO realization. It defines the possible output voltage levels

for MIMO converters. It incorporates all conditions mentioned above. In the following, a

detailed procedure is presented to show how to use (2.2) and (3.1) in designing and realizing

SC converters.

Suppose we have a system that requires two multiple output levels of 1.8V and 3.3V

from one input source, which is common in power management circuits. There are two

16

approaches to design a DC-DC converter for this application, either using two different DC-

DC converters or using the MIMO converter design concept. The latter has less component

counts and makes it the optimal choice for this application. The steps to design the MIMO

converter are as follows:

1. Defining how many inputs and outputs which allows us to find minimum limit for k.

2. Determining the maximum VCR using (2.2). Then, find new k and choose the highest

value for k amongst the two steps (1) and (2).

3. Assigning terminal weight following the conditions listed in section 2.3.

4. Connecting the voltage nodes to terminals based on their weight.

Let us use the example stated earlier to verify the design procedure steps. We have an input

voltage of 1V and we need to get 1.8V and 3.3V outputs. So, following the same design

steps:

1. n=1, m=2 which implies 1+2 ≤ k+2, and consequently, k need to be greater or equals

one.

2. Maximum VCR here is 3.3 which is close to 7/2, Using equation (2.2), k greater than or

equals four is required which corresponds to F6=8. Therefore, the circuit needs at least

four flying capacitors and six terminals (Figure 2.4) using FSC converter topology.

3. Terminal weights are assigned as following:

(a) w1 = Fk+2 = 8 and w6 = −1.
(b) w2 + w3 + w4 + w5 = −7, using a trial and error approach, w2 = −3, w3 = −2,
w4 = w5 = −1. Note that, these terminal weights should follow the Fibonacci

17

Figure 2.4: Circuit realization for FSC converter with dual outputs.

series as stated earlier. For convenience, the terminal weights are also highlighted

next to each terminal in Figure 2.4.

4. Output voltages relation is examined which helps in assigning the terminals. Detailed

discussion of this step is to follow.

In this example, we have Vo2 (cid:39) 2Vo1 as a design requirement (1.8V, 3.3V). Accordingly,

terminals need to be selected to guarantee this relation between the outputs. For instance, if

Vo1 is connected to the second terminal in this converter which has a weight of three (Figure

2.4), the second output needs to be connected to a terminal that provides a weight of six

which is impractical with this converter. In this case, Vo2 is connected to the third terminal

which has a weight of two and Vo1 can connected to any of the last three terminals because

they have same weights, w4, w5 and w6. For convenience, Vo1 is connected to the last

terminal. The input is connected to the first terminal, because positive VCRs are required.

Using (3.1), 8vg = 2vo2 + vo1 which implies 8vg = 5vo1 and 8vg = 3vo2.

Spice simulation was conducted for the example discussed above. Figure 2.4 shows the

circuit realization where four 15µF flying capacitors were used. The output capacitors are

18

C1SaSbSaVgVo1-38C2SbSaSb-2C3SaSbSa-1C4SbSaSb-1Sa-1Vo2Figure 2.5: Output voltage waveforms for dual outputs FSC converter.

kept high to maintain small output voltage ripple. The converter is operated using 50%

duty-ratio complementary non-overlapping gate signals, Sa and Sb, in open loop control.

These gate signals are provided using 500kHz pulse clock generator.

Ideal switches with

Rdson=10mΩ are used as power switches. The output voltage waveforms of 3.2V and 1.6V

are achieved using an input voltage of 1V. These converters outperform the single output

voltage converter in terms of complexity and component count.

2.4 Summary

In this chapter, two limitations in SC converter fundamental limit were discussed, and general

expressions to overcome these limitations have been proposed. The design procedure using

these expressions was introduced, which helps in designing SC converter for a variety of

applications. The MIMO FSC converter has been reported for the first time in literature

and the same concept can be extended to other SC topologies. Simulation results show an

agreement with the theoretical analysis.

19

00.20.40.60.81Time (ms)0123Vout (v)Vo1Vo2Chapter 3

SC Converter Synthesis

Synthesizing SC converter is a tedious and time consuming job and usually performed in an

ad-hoc approach. This chapter discuses existing works to overcome this problem. Then, we

propose an efficient algorithm to realize multi VCR FSC converter which is expected to be

useful for wide input applications1.

3.1 Related Work

Generally, there are many ways to configure the SC converter. For instance, if one FSC cell

of Figure 2.1 is considered, there are three possible combinations and it increases to seven

combinations if we add another FSC cell. Considering FSC converter with k flying capacitors,
there are Fk+2(Fk+2−1)+1 possible combinations. So, synthesizing the FSC converter is not

trivial and considered a major challenge in SC converter design compared to SI counterpart.

A generic SC structure has been studied in [16] but the proposed synthesizing technique

is not systematic and considers only one condition: whether the converter is designed for

step-down or step-up. Details of configuring specific VCR were not discussed. Different

approaches and algorithms were proposed to tackle the SC synthesis problem [14, 17–19]

which are focused to realize the FSC or its variants.

In [17], an interesting approach to configure SC converter was proposed by connecting all

1Part of the work presented in this chapter has been published in the proceedings of IEEE Workshop on

Control and Modeling for Power Electronics (COMPEL), 2017 [15].

20

capacitor negative plates to the ground in the charging phase, and hence it is known as All

Negative terminal connected To Zero topology (ANTZ), and then configuring the capacitors

to deliver the charge to the output based on optimal vector path from the ANTZ tree.

The ANTZ tree is composed of two main levels, the voltage source level, usually (Vin,Vin)

or simply (1,1), and flying capacitor levels. Each flying capacitor level is assigned a vector

based on the possible voltage appears in the flying capacitor terminals (Vci,Φ1,Vci,Φ2). Using

ANTZ, many SC topologies can be configured including Fibonacci topology. However, this

topology targets only step-up converters with non-fractional VCR and cannot be considered

as a generic model for SC converter synthesis. In addition, composing the ANTZ tree is quite

complicated especially for large number of flying capacitors and hence the design procedure

is not straightforward.

The algorithm found in [18] is quite simple but not effective in finding all configurations

for FSC converter. As an example, using this algorithm a V CR = 5/3 was found to be

realizable in only three configurations (Figure 6 in [18]). However, it can be realized in

four configurations using the proposed algorithm in this work, see Table 3.1. Details of

the proposed algorithm is discussed Also, assigning capacitor polarities needs to be done by

inspection which cannot be automated. In addition, the configuration table grows with the

increase in the number of flying capacitors which complicates the synthesis process.

The third algorithm we want to discuss here is proposed in [19]. It takes two steps to

synthesize the SC converter: (i) generating fundamental VCR using basic FSC converter and

(ii) summing these fundamental VCR to generate the desired one. This method works the

best with high VCR as it has less voltage stress than conventional FSC converter. However,

it fails to maintain the minimum number of flying capacitors required for specific VCR based

on the fundamental limit (2.2). Basically, it uses extra capacitors in the second part of the

21

Table 3.1: All possible realizations for FSC with VCR=5/3

Terminal
Weight

SPTT code

Final SPTT code
Number of Switches
Voltage nodes

t2
-2
0
2
2
2
2
0

t1
5
1
1
1
1
1
0
Vin Vout

t4
-1
2
0
2
0
0
0

t3
-1
2
0
0
2
0
0
gnd gnd Vout

t5
-1
2
2
0
0
2
0

system to perform the addition. Also, it can be used only for an integer step-up SC converters

like the ANTZ topology.

Recently, an algorithm to evaluate the SC converters is proposed in [20]. The evaluation

is based on charge-multiplier concept which will be discussed thoroughly in the following

chapters. However, their claim to be first attempt towards solving the SC synthesis is

misleading as we already discussed some prior works in the literature. Moreover, the work

in [20] shows only the ability to find charge-multiplier vectors for evaluation purposes, but

the synthesis approach is not clear. For example, there is no limits for the number of required

cells for specific VCR. Also, the algorithm targets a one-VCR applications and hence is not

practical for multi-VCR. Despite all of that, the generic nature of the approach can be

credited compared to topology-dependent approaches.

A canonical model was proposed in [14] to solve the realizability problem of FSC con-

verter. This model was successful to synthesize the FSC converter to achieve any VCR

including fractional and negative VCR. It proposed a gearbox circuit that configures the

converter such a way to be able to perform certain VCR. Each flying capacitor terminal is

assigned a weight based on Fibonacci series. Based on the terminal’s weight and specified

VCR, the gearbox circuit directs each terminal to specific voltage node (Vin, Vout, gnd) via

22

switches. However, it was developed only for a specific VCR and did not target multi VCRs

which is sometimes desirable. Here, we extends the canonical model to synthesize a truly

variable FSC converter by adjusting the gearbox switches. Details of the proposed algo-

rithm is discussed in the following section. Table 3.2 summarizes the synthesis techniques

with general comparison which concludes the effectiveness of the proposed algorithm in this

work where a general FSC that provide any VCR can be synthesized.

Table 3.2: Comparison of SC converter synthesis techniques

COMPEL’14 [14] TCAS-II’07 [17] COMEL’15 [18] EDSSC’09 [19] This work

Complexity level

Capacitor count

Variable VCR

Low

min.

No

High

min.

Yes

Avg.

min.

No

Avg.

High

No

Low

min.

Yes

3.2 Proposed FSC Synthesis

The canonical model proposed in [14] is shown in Figure 2.3. It consists of Fibonacci con-

verter cells and a gearbox to change the voltage conversion requirement. The synthesis

approach is approved mathematically, and hence it is used here as a basis to help synthe-

sizing variable FSC converter. The Makowski general SC converter starts with selecting

the minimum number of capacitors to achieve the maximum VCR by using equation (2.1).

This step determines the number of converter cells and the number of the terminals to the

gearbox.

The second step involves weight function assignments based on the VCR, this step is

done using equation (3.1). It is assumed that each path in the gearbox has a fictitious single

pole triple throw (SPTT) switch that connects the converter core to one of the three voltage

nodes (Vin, Vout, gnd) in the blue box. These terminals carry some weight depending on the

23

Fibonacci series, F = (1, 1, 2, 3, 5 . . . Fk+2), either in positive or negative values. Equation

(3.1) is a condition that ensures all weight function values are part of Fibonacci series as

follow:

k+2(cid:88)

i=2

w(t1) +

w(ti) = 0,

(3.1)

where w(t1)=Fk+2, w(tk+2)=−1 and w(tk)⊂[−Fk+1,−Fk.... − 2,−1,−1]. Finally, the

SPTT switch is assigned a code (0, 1 or 2) where 0, 1, and 2 means that the terminal is

connected to ground, Vin and Vout, respectively. These steps allow synthesizing the SC

Fibonacci converter for any VCR. Examples of positive and negative VCRs with thorough

explanations and mathematical proof can be found in [14]. So, this canonical converter is

very efficient in synthesizing SC converters but when considering a multi-VCR, an additional

step is required to ensure the optimal realization. In fact, the converter in Figure 2.3 may

be extended to other modular structural SC like serial-parallel, which was already covered

in [14] and called semi-canonical, ladder and other topologies. Also, this method can be

extended to multiple inputs and/or outputs which may find its way to different applications,

such as harvesting energy from multiple ambient sources. However, these extensions are out

of the scope of this work. The proposed algorithm to synthesize a variable FSC converter

takes the following steps:

1. All three steps mentioned in section 3.2;

(a) capacitor requirements.

(b) terminals weight function assignment.

(c) SPTT code assignment.

24

2. Unique/optimal realization assignment.

The first step including (a)-(c), is already discussed in chapter 3. For the second step,

the algorithm will search for the minimum number of state transitions in each realization

compared to a reference set by the code, usually the SPTT code that corresponds to the

maximum VCR. The number of state transitions here represents the number of switches in

the gearbox circuit. Note that, these switches configure the SC converter in such a way that

it can achieve all pre-assigned VCRs and hence they are considered the main attribute to

the complexity level of the controller.

The whole algorithm flow chart is depicted in Figure 3.1. In this code, the user needs to

specify the VCRs and degree of freedom, i.e resolution. The code will generate the minimum

number of capacitors and terminals for each VCR using (2.2). Then, the weight functions

are next to be determined. After that, the SPTT code for each realization is found, and an

SPTT code reference is set by the code which is an important step to ensure all VCRs are

realizable with the final solution. Next, the code allows to compare all SPTT codes to the

reference and return SPTT codes that have a minimum number of state transitions. Note

that, a state transition occurs whenever there is a change in the SPTT code for each terminal

from (0 to 1), (0 to 2), (1 to 2), (2 to 1), or (2 to 0). The final SPTT code is found to be

the optimal realization in terms of number of switches in the gearbox. The SPTT code is
m× t matrix where m represents the number of realizable VCR and t represents the number

of terminals in the gearbox.

In some cases, the VCR required by the used can be relaxed, i.e 0.75 is assigned instead

of 0.76, which allows realizing the converter by less number of capacitors. Therefore, to

account for this degree of freedom, the resolution feature is added.

It simply allows the

algorithm to search for optimal VCRs that have the least number of capacitors. For exam-

25

Figure 3.1: Flow chart for the proposed algorithm.

26

NoNow(t0)=M, w(tk+1)=-1use Eq(3.1) to find all weights  M=Fk+2k= number of capacitors Apply Eq(2.2) to test -VCRIs M=Fib number?M=Q M=PIs Q<PInitialization: VCRs=P/Q, resolutionFinal solution (Number of FSC cells and terminal assignment)M=M+1 Are all VCRs found? w(tn)=Fk+1 n=n+1NoFind SPTT for all VCRs, SPTT reference=max{VCR} representation    Is number of transition minimum?Delete the row, r=r+1NoYesYesYesYesLoad MNoNow(t0)=M, w(tk+1)=-1use Eq(3.1) to find all weights  M=Fk+2k= number of capacitors Apply Eq(2.2) to test -VCRIs M=Fib number?M=Q M=PIs Q<PInitialization: VCRs=P/Q, resolutionFinal solution (Number of FSC cells and terminal assignment)M=M+1 Are all VCRs found? w(tn)=Fk+1 n=n+1NoFind SPTT for all VCRs, SPTT reference=max{VCR} representation    Is number of transition minimum?Delete the row, r=r+1NoYesYesYesYesLoad MRelax VCR based on resolutionple, for VCR=0.76 and zero resolution, the code will assign seven flying capacitors because

0.76=19/20 in the simplest form. However, a resolution of 0.01 would relax the VCR to 0.75

and hence, only three flying capacitors are assigned. In fact, this feature allows finding all

VCRs and helps the user to optimize the number of flying capacitors based on the specifica-

tions. The algorithm is implemented in MATLAB, and is made available as an open-source

tool, called FSC Synthesizer. Using this open-source tool, the user can realize any VCR

for FSC more efficiently, saving the time and effort required to go through trial and error

route. In the following sections, some cases are presented to illustrate the workflow of FSC

Synthesizer.

3.3 Case Study and Results

In this section, three cases will be discussed. These cases cover all types of FSC converter,

single VCR and variable VCR. The last example shows the effectiveness of the proposed

approach to simplify the controller circuit of an FSC converter.

3.3.1 FSC Converter with VCR=30

The first example is to implement an FSC that has a single VCR equals to 30, which is one

of the cases discussed in [14]. Using the proposed method, all solutions are validated and

shown in Table 3.3. In addition, the proposed method selected the optimized SPTT code as

shown in the Table which shows where the terminals are need to be connected. It also can

be seen that no gearbox switches are needed which is expected in single VCR FSC converter.

27

Table 3.3: All possible realizations for VCR=30/1

Terminal
Weight

SPTT code

Final SPTT code
Number of Switches
Voltage nodes

t1
34
1
1
1
1
1
1
1
1
1
1
0
Vin

t3
-8
0
0
0
2
0
0
0
0
0
0
0

t2
-13
0
2
2
2
0
0
0
0
0
0
0
gnd gnd gnd Vin

t4
-5
0
0
2
0
0
0
0
0
0
0
0

t5
-3
0
0
0
1
1
1
1
1
1
1
0

t7
-1
2
1
1
0
2
0
2
1
1
0
0

t6
-2
1
1
1
0
0
0
0
0
0
0
0
gnd gnd Vin Vout

t8
-1
1
2
1
2
0
1
1
0
2
1
0

t9
-1
1
1
2
1
1
2
0
2
0
2
0

3.3.2 Variable FSC Converter with VCR=5, 4, 3, 5/2

The second example considers the application discussed in chapter 1, where a variable FSC

is required. As mentioned earlier, the miniaturized solar cell has a typical open voltage of
0.3∼0.6V and we want a regulated output voltage of 1.5V. Therefore, the VCR=5, 4, 3 and

5/2 are required. These VCRs are fed to the proposed code in section 3.2 to get the optimal

circuit realization. This application needs only three flying capacitors using FSC topology.

The SPTT gearbox code and number of switches are presented in Table 3.4. In this example,

terminals 1, 2, and 5 preserve only one value along all circuit states which means no switch

is required and the terminal is connected directly to its corresponding voltage node. In third

terminal, however, the rows are toggling between zero and one which means two switches

are required. For the forth terminal, the rows are changed between all voltage nodes which

means three switches are required. Also by using FSC synthesizer as shown Figure 3.2, this

converter requires three flying capacitors and five transistors for the gearbox control. Note

that, due to the presence of all VCR in their simplest form, a zero resolution is chosen. The

28

SC converter realization is drawn in Figure 3.3. Note that, Sg1 − Sg5 transistors2 configure

the SC converter to change VCR if needed. Also, they are the only transistors that are

controlled in closed loop, all other switches are driven by simple two complementary non-

overlapping clocks that are enabled when the converter starts up. This converter will be

used in next chapters as a case study to implement variable FSC in both discrete and IC

implementation.

Figure 3.2: Terminal assignments using FSC Synthesizer for variable FSC with VCR=5, 3,
4, 5/2.

3.3.3 Variable FSC Converter with VCR=1, 3/4, 2/3

The complexity of the converter control circuit is also another issue in SC converter design

and can be related to the number of driving signals generated in close loop. Therefore, the

less number of closed loop driving signals the simpler is the control circuit. We claim that

2Note that, Mx and Sx are exchangeably used throughout this dissertation.

29

Table 3.4: Optimized realization for variable FSC with VCR=5, 4, 3, 5/2

Terminal
Weight

Final SPTT code

Number of Switches

t1
5
1
1
1
1
0

t2
-2
0
0
0
0
0

Voltage nodes

Vin

gnd

t3
-1
0
0
1
0
2

Vin
gnd

t4
-1
0
1
1
2
3
Vin
Vout
gnd

t5
-1
2
2
2
2
0

Gain
NA
5
4
3
5/2
NA

Vout NA

Figure 3.3: Variable FSC with VCR=5, 3, 4, 5/2.

30

C1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4Sg5nMOSpMOSTGby using the proposed algorithm, the simplest controller circuit is achieved. We illustrate

by comparing the control circuit of a buck converter that has three VCR (1, 3/4, 2/3) imple-

mented using our method and the same converter in the literature implemented in ad-hoc

way [21].

Implementing this buck converter using our methodology resulted in a circuit

shown in Figure 3.4(a) with an LUT shown in Table 3.5. In [21], three capacitors and 13

switches were used to realize a buck converter with V CR = (1, 3/4, 2/3) with an LUT control

signals shown in Table 3.6, their circuit realization is redrawn in Figure 3.4(b). With a close

look to the switch signals in Tables 3.5 and 3.6, one can see that ten switches (all except

M1, M3 and M5) of the circuit in Figure 3.4(b) need to be controlled in a closed-loop, i.e.

they are not just a simple on-off (two states) control. On the other hand and for the same
V CR, only four switches ( Mg1 − Mg4) are controlled adaptively as illustrated in Table 3.5

using the proposed method. Not only that, but also each two switches among the four are

complementary, Mg1 = ¯Mg2 and Mg3 = ¯Mg4. Thus, only two signals are required to control

the gearbox switches. Other switches, M1−10, are controlled by the non-overlapping clocks

(φ1 and φ2) directly. This greatly simplifies the converter control unit.

Table 3.5: LUT for converter synthesized by proposed method

1
V CR
M1−10(odd)
φ1
M1−10(even) φ2
0
Mg1
1
Mg2
1
Mg3
0
Mg4

3/4
φ1
φ2
1
0
1
0

2/3
φ1
φ2
1
0
0
1

31

Figure 3.4: Buck SC converter with VCR=1, 3/4, 2/3: (a) FSC topology using proposed
method. (b) Converter proposed in [21].

32

C1C2C3Mg4Mg3M9Mg2Mg1M4M6M1M2M3M7M10M8M5VinCoutC1C2C3M11M12M6M8M9M3M2M4M13M10VinCoutM7M5M1(a)(b)Table 3.6: LUT for converter in [21]

V CR 1
φ1
M1
0
M2
φ2
M3
1
M4
M5
φ1
0
M6
φ2
M7
1
M8
0
M9
0
M10
0
M11
0
M12
0
M13

3/4
φ1
φ1
φ2
0
φ1
φ1
0
0
φ2
φ1
φ1
φ2
φ2

2/3
φ1
φ1
φ2
0
φ1
φ1
0
φ2
φ2
φ2
0
0
0

3.4 Summary

The SC converter synthesis problem is introduced and solutions are discussed. The proposed

algorithm achieves an optimal SC realization for mulit-VCR in almost no time which makes

it desirable as a fast tool to synthesize SC converters. As practical examples, three cases

were presented. The proposed method also simplifies the converter control by decreasing

the number of close loop driving signals. A comparison between the control method for one

converter found in literature and the other synthesized in the proposed method is discussed.

For the same number of VCRs, it is found that the control of the proposed one is much easier

compared to the other.

33

Chapter 4

Reconfigurable FSC Converter:

Discrete Design

This chapter aims to study the design and implementation of discrete FSC converter. Al-

though multi-phase FSC converters provide more VCR as discussed earlier, the two-phases

FSC are to be considered in this work due their simplicity in design and control. The discrete

differs from the IC integration in terms of power loss analysis and components selection.

4.1 Theoretical Analysis

The SC converter equivalent circuit is shown in Figure 4.1. It consists of an ideal conver-

sion term (n = V CR), and equivalent impedance (Req) which models all switch turn-on

resistances and charge redistribution in the flying capacitors. The output voltage can be

calculated using the following equation:

Vout = nVin

1
Req
RL

1 +

.

(4.1)

34

Figure 4.1: SC converter steady state model.

4.1.1 Converter Equivalent Resistance

There are enormous published works trying to improve the SC models. Some models are

summarized and compared in recent work [22]. The model proposed in [23] will be utilized

in this paper due to its generic nature and easy implementation for complex SC topologies

like FSC. Also, it provides useful insight to component sizing and converter optimization.

There are two switching limits for Req as shown in Figure 4.2: slow switching limit (SSL)

and fast switching limit (FSL) [23–26]. The slow switching occurs when the switching period

(Ts) is less than the flying capacitor time constant (τc = RCf ly), in other words fs < fc

as illustrated in Figure 4.2, and the switches in SSL region can be assumed to be ideal, i.e

Rron = 0 and Rrof f = ∞. On the other hand, the fast switching spans the region where fs is
much higher than fc, and Req is minimized. In fact, RF SL, which is the converter impedance

at FSL, is considered the minimum impedance value a converter can achieve. Nevertheless,

the converter is usually operated around fc or less to avoid high switching loss and achieve

smooth charge transfer. In fast switching mode, the flying capacitor can be modeled as an

ideal voltage source, which implies there is no charge loss in charging and discharging the

capacitor or so-called charge redistribution loss [16]. To get an optimal equivalent resistance,

both switching regions need to be considered. The converter impedance at SSL and FSL,

35

ReqRL1:nVinVoutRSSL and RF SL, and the total converter equivalent resistor (Req) can be calculated as

following:

k(cid:88)

i=1

(a2
)
ci
Cifs

,

RSSL =

m(cid:88)

j=1

(a2
rj

)Ronj
Dj

,

R2
SSL + R2

F SL,

RF SL =

(cid:113)

Req =

(4.2)

(4.3)

(4.4)

where where fs is the switching frequency, Ron is the switch turn-on resistance, D is the duty

ratio of the gate signals which is normally equals 0.5, k is the number of flying capacitors and

m is the number of switches. The aci and arj are the charge multiplier of the ith capacitor
and the jth switch respectively.

Figure 4.2: SC converter equivalent impedance vs switching frequency.

36

SSLFSLfcFs (Hz)Req  (Ω)AsymptoticRealThe charge multiplier vectors, ar and ac, relate the charge flows into the capacitor and

switch to the output charge, meaning arj = qrj /qout and aci = qci/qout. The charge vectors

are very useful in converter optimization as will be discussed later. To find the charge vectors,

sub-vectors a1

r,c and a2

r,c which define the charge vectors in each phase, need to be determined

by drawing the converter in each state and finding the charge in each element. Due to the

charge conservation, the following equations; q1

out + q2

out = 1 and a1

rj ,ci

+ a2

rj ,ci

= 0, helps

identifying the charge quantity in each phase. Then, aci can be found by taking the absolute
value of a1
ci

. Likewise, arj can be determined by taking the corresponding nonzero element

in a1
rj

and a2
rj

.

Figure 4.3: FSC converter with VCR=3.

To follow with theoretical analysis, an example of FSC converter with VCR equals three

is shown in Figure 4.3. The converter has two states as shown in Figure 4.4, which will be

used to compose a1

c,r, a2

c,r and ac,r as following:

 =

 1 −1


−1

1

a1

c1
a2
c1

a1
c2
a2
c2

37

C1VinC2VoutM7M4M6M1M3M5M2Figure 4.4: FSC converter with VCR=3: (a) state I, Sa=1, Sb=0. (b) state II, Sa=0 Sb=1.

(cid:20)

=⇒ ac =

1 1

(cid:21)
 =

a1

r1
a2
r1

a1
r2
a2
r2

a1
r3
a2
r3

a1
r4
a2
r4

a1
r5
a2
r5

=⇒ ar =

a1
r6
a2
r6

a1
r7
a2
r7

(cid:20)

(4.5)

2 0 1 0 1 0 1

(cid:21)

0 1 0 1 0 1 0

2 1 1 1 1 1 1

(4.6)

In this work, a blocking voltage vector is also introduced, avc,r . This vector defines

the blocking and rated voltages for the capacitors and switches with respect to the output

voltage, avc,r = Vc,r/Vout. To do this, we first find Vc,r in terms of Vin, and then divide

by VCR to get Vc,r as a function of Vout. Also, we could find Vc,r in terms of Vout in

38

C1VinC2VoutqoutM4M62qoutM2qoutC1VinC2VoutM5M1qoutM3M7qout(a)(b)one step by reversing the analysis. Note that, all previous analysis is related to the output

voltage and charge. Therefore, we need Vc,r to be as a function of Vout to be consistent

and get a meaningful results that helps analyzing and optimizing the converter. For the

switches, the blocking voltage is the maximum voltage which the switch can withstands

without breaking out, and usually determined during the off state. Likewise, the blocking

voltage for the capacitor is defined to be the maximum voltage across the capacitor in both

states. Defining the blocking voltage vectors is very useful in power analysis calculation

which will be discussed in the next section. As an example, let us consider the converter

discussed earlier (Figure 4.3), avc,r can be found as following:

a1

vc1
a2
vc1

 =
(cid:20)

1/3 2/3

(cid:21)

1/3 2/3

a1
vc2
a2
vc2

=⇒ avc =

a1

vr1
a2
vr1

a1
vr2
a2
vr2

a1
vr3
a2
vr3

a1
vr4
a2
vr4

a1
vr5
a2
vr5

a1
vr6
a2
vr6

a1
vr7
a2
vr7

(cid:20)

=⇒ avr =

1/3 2/3

 =

 0

1/3

(4.7)



1/3

1/3

0

2/3

0

1/3

0

0

1/3

2/3

0

0

(cid:21)

1/3 1/3 1/3 2/3 2/3 1/3 1/3

(4.8)

Finally and to have a concrete theoretical analysis, let us study the optimized converter

expressions. The optimization procedure proposed in [24] explicitly derived the expressions
of optimized R∗

F SL based on ac and ar respectively. This optimization minimizes

SSL and R∗

39

Ploss by sizing the flying capacitor and transistors according to ac and ar vectors. Therefore,
R∗
SSL and R∗

F SL can be calculated as following:

k(cid:88)

i=1

R∗
SSL =

1

Ctfs

(

aci)2,

m(cid:88)

arj )2,

R∗
F SL = 2Ront(

j=1

(4.9)

(4.10)

(4.11)

(4.12)

Ct,

k ack

aci(cid:80)
(cid:80)

C∗
i =

R∗
onj

=

m arm
arj

Ront,

where Ct is the total flying capacitance in the converter and Ront is the total turn-on

resistances of all switches in the converter. For discrete implementation, (4.3) and (4.9) will

be used to quantify Req because the switch turn-on resistance is fixed once the transistor

is selected and hence it can not be optimized. The flying capacitors are sized using (4.11).

The av and ar vectors are useful for selecting suitable capacitor and switch rating.

4.1.2 Power Loss Analysis

The efficiency in SC converters can be categorized into two parts: the voltage conversion effi-

ciency (VCE) and power conversion efficiency (PCE or commonly η). They can be quantified

by the following equations:

V CE =

Vout
V T
out

,

40

(4.13)

η =

Pout

Pout + Ploss

,

(4.14)

where VT

out is the target output voltage, and equals nVin seen across the secondary side of the

ideal transformer in the average state model (Figure 4.1). Excluding the bottom-plate charge

loss, the power losses (Ploss) in low power SC converters are due to two main factors [27] as

following:

1. Linear losses (Plin): these include the conduction loss and charge redistribution loss in

the flying capacitors. Due to their negligible effect, effective series resistor (ESR) of the

flying capacitors, trace impedance are usually ignored. Then, Plin can be calculated

as following:

Plin = I2

outReq,

(cid:118)(cid:117)(cid:117)(cid:117)(cid:116)(cid:34)(cid:0)(cid:80)k

i=1 aci
Ctfs

(4.15)

(cid:35)2

(a2
rj

)Ronj

.

(4.16)

(cid:1)2

(cid:35)2

(cid:34)

+

2

m(cid:88)

j=1

=⇒ Plin = I2

out

2. Switching losses (Psw): these primarily include gate driving loss, diode reverse recovery

loss, switch turn-on and turn-off losses. The diode reserved recovery will be neglected

due to the low voltage across the switches. During turn-on, the SC converter is assumed

to be driving a soft load and it drives a hard load in the turn-off interval, Figure 4.5.

Therefore, Pon and Pof f can be found using the following:

Pon =

1
2

IdsVdsfstf ,

41

(4.17)

Pof f =

1
6

IdsVdsfstr.

(4.18)

Then by using the charge and blocking voltage vectors, (4.17) and (4.18) can be mod-

ified to the following:

Pon =

ariavi

2

IoutVoutfstf ,

Pof f =

ariavi

6

IoutVoutfstr.

(4.19)

(4.20)

=⇒ Psw =

m(cid:88)

j=1

Qgj Vgsj fs +

(cid:16) tf

2

(cid:17)

+

tr
6

IoutVoutfs

m(cid:88)

j=1

arj avj .

(4.21)

Figure 4.5: Gate switching intervals.

The switching power loss can be further optimized by employing soft-charging, or com-

monly referred to as zero-voltage-switching (ZVS) or zero-current-switching (ZCS), where

42

Sa,bVdsIdstd(on)tftd(off)trhard loadsoft loadthe switch is allowed to turn-on when Vds reach zero or close to zero and the same for the turn

off state with ZCS. The soft-switching can be achieved in resonant SC converters which has

inductors in series with the flying capacitors. However, this technique limits the switching

frequency to the resonant frequency, and it is undesirable to implement in converters target-

ing IC integration. The other way is to allow high output voltage ripple of the SC converter

by reducing Cout (ideally Cout = 0) and then the output voltage ripple is controlled using

a buck converter [28]. Nevertheless, in both techniques magnetic element is required which

contradicts with the motivation for implementing purely capacitive converters. In addition,

Psw is very small compared to Plin in low power SC converters. Therefore, the soft-switching

is not considered in this work.

4.1.3 Output Regulation Methods

The SC converters suffer from poor output regulation which considered the second main

challenge. Obviously from (4.1) and (4.13), there are two approaches to achieve relatively

good output regulation: intrinsic and extrinsic control methods. Many of the control meth-

ods are discussed and compared in [26]. In general, intrinsic control involves changing Req

through the parameters seen in (4.4), which consequently fine tunes the output voltage.

These methods include duty ratio modulation (D), capacitor modulation (Ct), transistor

width modulation (Ron and switching frequency modulation (fs). Although RF SL is highly

dependent on the duty ratio, this technique resulted in very minimal effect in the total

power profile. In addition, 50% duty ratio complementary driving signals requires a simple

control circuit, compared to adaptive duty ratio signals, hence this technique is excluded.

Component size modulation including Ct and Ron can be utilized, but in price of adding

more components which increases the switching losses due to added switches that performs

43

Table 4.1: Variable FSC converter design specifications

Parameter Vin
Value

Vout

V CR

0.3-0.6 V 1.5 V 5, 4, 3, 5/2

Pout
50 mW

modulation, and it also decreases the power density. However, these techniques can be em-

ployed in IC with careful size matching and efficient control. The frequency modulation is

commonly used to regulate the output voltage in SC converters.

The extrinsic control method involves changing the VCR, i.e adaptive VCR, which of-

fers a course tuning improving the VCE. Due to the ability of SC converters to provides

multiple VCRs, the adaptive VCR method is quite effective to regulate the output voltage

for applications that encounter wide input voltage range [29, 30]. For example, in FSC con-

verter with two flying capacitors, seven possible VCR can be achieved. By enabling the

adaptive VCR, this converter can be used to maintain relatively good VCE for wide input

voltage range. We believe the adaptive VCR is an efficient regulation method compared to

the intrinsic techniques for wide input range applications. However, there is a trade off by

adding more switches to control transitions between different VCR. The addition of these

transistors would increase Plin and Psw. Nevertheless, this technique is adopted in this work

and careful optimization to these transition switches is considered.

4.2 Converter Design

In this section, we demonstrate the use of the theoretical analysis discussed in previous

section by designing a variable FSC converter. The summary of the design specifications are

listed in Table 4.1. This converter can be synthesized using the proposed algorithm discussed

in chapter 3. For convenience, the circuit realization is redrawn in Figure 4.6.

44

Figure 4.6: Variable FSC with VCR=5, 3, 4, 5/2.

4.2.1 Power Loss Calculation

Referring to the circuit states of the converter in Figure 4.7 and following the same procedure

in section 4.1.2, ac, ar and av can be found as following:



acV CR=5

acV CR=4

acV CR=3

 =



2

2

2



1

1

1

1

1

1

acV CR=5/2

1 1/2 1/2

45



arV CR=5

arV CR=4

arV CR=3

arV CR=5/2

 =



3

3

2

2 2

2 2

2 2

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

0

1

1

1

0

0

0 1

0 1

1 0

0

0

0

3/2 1 1 1/2 1/2 1/2 1/2 1/2 1/2 1/2 0 1/2 0 0 1/2



C1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4Sg5nMOSpMOSTGFigure 4.7: All operation states for proposed FSC: (a) VCR=5, (b) VCR=4, (c) VCR=3,
and (d) VCR=5/2.

46

C1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqoutC1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqout3qout(a)C1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqoutC1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqout3qout(b)C1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqoutC1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqout3qout(c)C1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqoutC1S1S2S3C2S4S5S6C3CoutVoutSoutVinS8S9S7Sg2Sg1Sg3Sg4qout2qoutqout3qout(d)Sg5Sg5
 =




avrV CR=5

avrV CR=4

avrV CR=3
avrV CR=5/2

 =





1/5 2/5 3/5

1/4 1/2 1/2

1/3 2/3 1/3

2/5 2/5 1/5

avcV CR=5

avcV CR=4

avcV CR=3
avcV CR=5/2

1
5

1
4

1
3

2
5

1
5

1
4

1
3

2
5

1
5

1
4

1
3

2
5

2
5

1
2

2
3

4
5

2
5

1
2

1
3

4
5

1
5

0

0

2
5

3
5

1
2

1
3

1
5

3
5

1
2

1
3

1
5

2
5

1
2

2
3

1
5

2
5

1
2

2
3

1
5

1
5

0

0

2
5

0

1
4

1
3

0

1
5

1
4

0

3
5



0 1

0 1

1
3

2
3

1 0

In this converter, we will consider the case of highest VCR which is associated with the

(4.22)



worst case in terms of Ploss. Therefore, the following vectors is used:

 ac

 =

 2

avc

1

1

1/5 2/5 3/5



 =

 ar

avr

 3

2

2

1

1

1

1

1

1

1

0

1

0

1 0

1/5 1/5 1/5 2/5 2/5 1/5 3/5 3/5 2/5 2/5 1/5 0 1/5 0 1

(4.23)
Due to the zero elements of the gearbox transistor , Sg1 − Sg5, their switching losses

will result in zero loss when we multiply ar by avr , which increases the error in calculation.

Thus, a way to account for their loss is need. From Figure 4.7, only two gearbox switches

are observed to on at each VCR. Besides that, they are turned on in the transition between

different VCR only. Therefore, we could account for their loss by considering the maximum

47

value for their charge and voltage representations. Referring to the matrices, one can notice

that the maximum values for Sg4 with ar=1 at VCR=(4,5) and avr =1 at VCR=5/2 states.

As a result, gearbox control switch losses carry a weight of two, which is considered a worst

case because usually the gearbox is not switching at the rate of system clocking signal. Then

by using (4.16) and (4.21), the Ploss can be found as following:

(cid:115)

256
C2
t f 2
s

Ploss = I2
out

+ 2352R2

on + 12QgVgsfs + 3.2

(cid:16)

(cid:17)

tf +

tr
3

IoutVoutfs.

(4.24)

4.2.2 Components Selection

Using (4.11), the flying capacitance is sized as following, C1 = 1

2 Ct and C2 = C3 = 1

4 Ct.

Ct = 9.1µF have been selected based on rigorous simulation runs using genetic algorithm

to find the global minimum of Ploss in each run. Different commercial switches have been

examined and compared to select the optimal switch with highest efficiency to be use as a

power switch in the variable FSC converter. Table 4.2 summarizes the important parameters

of the chosen commercial switches. Various MATLAB simulations using real coded genetic

algorithm GA developed in [31] have been conducted. Figure 4.8 shows the Ploss verses

fs for all switches. Then and once optimal switching frequency for each switch is selected,

the efficiency profiles are plotted against the output power, Figure 4.9. Obviously, SW1 has

superior performance amongst all switches scoring 70% efficiency for 50mW output power at

90.1kHz switching frequency. Therefore, SW1 (BSS816NW from Infineon) has been selected

as a power switch for this variable FSC converter and the switching frequency is rounded

to 100kHz for easier manipulation and implementation. A complete design flow chart of

48

Table 4.2: Different chimerical switches with their specifications.

No.

Switch

Manufacturer

SW1 BSS816NW Infineon
SW2 SSM3K324R Toshiba
SW3 STR2N2VH5 ST
SW4 PSMN4R5
SW5 RYM002N05 ROHM

NXP

Qg
(nC)
0.6
2.2
4.6
20.5
1.3

Rds
(mΩ)
240
110
40
6.1
9000

tr
(ns)
9
9
14.4
18.7
8

tf
(ns)
2.2
9.5
4
8.75
43

variable FSC converter using discrete components is developed and can be found in Figure

4.10.

It combines all the steps discussed in chapter 3 and 4 providing a design guide to

implement FSC converter.

Figure 4.8: Ploss verses switching frequency for different switches.

4.2.3 Control Circuit

The LUT for gate signals control is shown in Table 4.3 (same as Table 3.5). The control
circuit provides two types of signals: (1) main clock signals Sa and Sb to drive S1 − S10
and (2) gearbox closed loop control signals for driving Sg1 − Sg5
1. The clock signals can be

1Hereafter, Sr means the driving signal of rth switch (Mr). Sa and Sb are the same as Φ1 and Φ2.

49

104105106fs (Hz)100Ploss (W)SW1SW2SW3SW4SW5Figure 4.9: Efficiency for different loads.

generated using simple ring oscillator (RO) where odd number of logic inverters are connected

together as shown in Figure 4.11. The frequency of oscillation can be calculated using the

following equation:

fos =

1

2nroτd

,

(4.25)

where nro is the number of logic inverters, nro = 3, 5, 7, ...(odd). τd is the logic inverter

propagation delay and it is usually reported in the data sheet of commercial logic inverters2.

For low fos using RO, usually higher nro and/or τd is required. Therefore, C(ro) is placed

in each stage to increase τd as shown in Figure 4.11. Then, non-overlapping clocks can be

generated using the circuit in Figure 4.11, where Cd is added to control the dead-time period.

Using Karnaugh’s map [32] and the LUT shown in Table 4.3, the expressions of these

signals can be found as following: Sg1 = ¯AB = Sg2, Sg3 = ¯A(B + C), Sg4 = ¯A ¯B ¯C and

Sg5 = A, where A, B and C are the comparators output signals. The complete control circuit

2The inverter equivalent resistance seen at the output can be approximated by calculating the frequency
generated from minimum RO of three stages for two or three different C(ro) values for comparison purposes.

50

00.050.10.15Pout (W)020406080 (%)SW1SW2SW3SW4SW5Figure 4.10: Variable FSC converter design flow chart using discrete components.

51

Specifications: Vin, Vout,  Vout, Pout, ηUsing VCR defined in step 1  Apply FSC synthesis procedure in Chapter 3Define ar, ac, av vectors of the worst case using procedure in Chapter 4Derive Ploss (Iout, Vout, Ct, fs, Qg, Vgs, tr, tf) function{Note, flying capacitors are optimized here using (4.11)}Using any numerical method like GA, for specified Pout find global minima of Ploss (Ct, fs, Qg, Vgs, tr, tf)Is η acceptable?Select another switchFinal Circuit realization with components sizesNoYesFigure 4.11: Ring oscillator with non-overlapping clocks generator.

Table 4.3: LUT for variable FSC converter.

V CR
S1−10(odd)
S1−10(even)
Sg1
Sg2
Sg3
Sg4
Sg5

5
Sa
Sb
of f
on
of f
on
of f

4
Sa
Sb
of f
on
on
of f
of f

3
Sa
Sb
on
of f
on
of f
of f

5/2
Sa
Sb
of f
on
of f
of f
on

52

Cd   CdCroCroCroSaSbNon-overlapping clocks generatorfosthat generates Sg1 − Sg5 is depicted in Figure 4.12. We use dual ultra-power comparators

(LTC6702 from Linear Inc.).

Figure 4.12: The gearbox signals control circuit.

Figure 4.13: Variable FSC converter PCB.

53

R1R2VinR3R4VDDSg1Sg2Sg3Sg4Sg5ABC4.3 Results

A prototype pf the designed variable FSC is implemented using two layers printed circuit
board (PCB), see Figure 4.13. It occupies an area of (36 ∗ 48mm2). We implement the ring

oscillator using 74HC04 inverters and Cro=15pF. Regarding the dead-time, and as shown in

Table 4.2, SW1 needs only a few ns to switch off. For the non-overlapping clock generator

we used, we got around 125ns dead-time but when they are loaded by the switches, the

clocks experience overlapping period which is something we want to avoid. To overcome this

problem, the delay time in the non-overlapping circuit is increased by adding 1.5nF delay

capacitance.

Figure 4.14: Vout response to Vin for variable FSC converter.

The output voltage response for different number of input voltages is drawn in Figure

4.14. It is clear that the output voltage response is as expected in the theory and synthesis

process. The mean values of the output voltage are 1.51V and 1.46V for the simulation and

experiment, respectively. Therefore, the output voltage mean value deviates a little from the

nominal value (1.5), with an error of almost 2.67% for the experiment results. Overall, the

54

0.350.40.450.50.550.6Vin (V)11.21.41.61.82Vout (v) TheoreticalSimulationExperimentexperiment and simulation results show an agreement with the theory.

4.4 Summary

In this chapter, we discussed the design and analysis of discrete variable FSC converter. We

developed a design procedure flow that helps designing an efficient FSC converter. A variable

FSC prototype is implemented using discrete components to show the functionality of the

proposed converter. Experiment results show an agreement to the theory with an error of

2.46% in the output voltage mean value over an input voltage range of 0.3-0.6V.

55

Chapter 5

Integrated FSC Converter

In this chapter, an overview on integrated FSC converter (IC FSC) challenges and design is

discussed. The proposed converter is designed and simulated using standard 0.5µm CMOS1

technology provided by ON Semiconductor foundry, commonly called C5 [33]. Details about

this technology and layout implementation of the devices are to be discussed in this chapter.

The details of the startup circuit and controller design, layout and measurement results are

presented in chapter 6.

5.1 On-chip Devices

In this section, the on-chip implementation of the switch, the solar cell and the capacitor is

discussed. The later two devices comprised most of the die area. Optimizing their area is the

road to high power density circuits. The switch area can be optimized to achieve minimum

Ron, which helps improving efficiency performance. In this work, however, solar cells are not

considered for same die integration. Nevertheless, remarks on IC solar cells are presented to

pave the road for an ultimate smart solar cell2.

1Complementary Metal Oxide Semiconductor
2Smart solar cell concept is introduced in chapter 1.

56

5.1.1 Switch Implementation

The switch can be implemented using PMOS, NMOS or transmission gate (TG) which is

a parallel combination of PMOS and NMOS. The PMOS and NMOS transistor structures

in CMOS are shown in Figure 5.1. They have four terminals: drain (D), source (S), gate

(G) and body (B). For simplicity, we can assume that these transistors are working in triode

region, also called ohmic or linear region in some references. The switch-on resistance is the

channel resistance between D and S terminals of the transistor, and can be quantified as

following:

Ronp =

Ronn =

1

µpCox( W

L )p(Vsg − |Vtp|)

,

1

µnCox( W

L )n(Vgs − Vtn)

,

(5.1)

(5.2)

where µp and µn are the mobility of electrons, cox is the oxide capacitance per unit area,

Vtp and Vtn are the threshold voltages of PMOS and NMOS, and they are all technology

dependent parameters. W and L are the gate width and length of the transistor, respectively.

In case of using TG, RonT G = Ronp//Ronn.

Figure 5.1: CMOS structure.

57

n+p+p+n+p+n+p-substraten-wellBDGSBDGSNMOSPMOSVDDUsually, the body terminal of NMOS is connected to lowest potential to avoid latch-up

that may occurs due to the connectivity of internal diode between B and S/D. For PMOS

and for the same latch-up concern, the body terminal is commonly connected to the highest

potential. In most cases where their is freedom in body connection, source terminals are

usually connected to body terminals. However, there are some cases in n-well process3

where the source terminal can not be connected to gnd or VDD like most of the flying

switches in SC converter structure. There are some non-idealities associated with MOSFET

when it is used as a flying switch [34], summarized as follows:

1. Charge injection: when the MOSFET is turned on, some charge (Qch) accumulates

in the transistor channel formed between the source and the drain, and the charge is

traped with no path to discharge. During MOSFET switch-off state, Qch is injected

to the load capacitor (C). Consequently, the voltage across the capacitor decreases to

maintain a charge balance. The charge injection phenomena can be studied using the

following equations:

Qch = W LCox(Vc − Vx),

W LCox(Vc − Vx)

C

.

Vc =

(5.3)

(5.4)

The charge injection can be solved by placing a dummy switch in the charging path

with an inverted gate signal to the main switch. This dummy switch prevents the

charge injection by soaking the charge during turning off the main switch. Another

3CMOS can be fabricated using n-well, p-well or triple-well process. C5 used in this work is an n-well

process.

58

solution is to use TG switch where two channels of PMOS and NMOS are available at

the same time, and they work as a reservoir that maintains Qch from leaking to either

terminals. A third option can be considered by keeping the ratio of Cch to C very low

meaning W LCox << C.

2. The turn on resistance of the flying switch depends on the input and/or output voltages

if they are connected to the source terminal. In fact, when Vgs approaches Vt, the Ron

increases drastically as described by (5.2) and (5.1). On the other hand, using TG,

this effect can be minimized due to the parallel combination of Rn and Rp. Second

approach to tackle this problem is to apply higher gate voltage to maintain enough

potential difference between Vgs and Vt.

3. Body effect: the threshold voltage of a transistor (Vt) is proportional to the the change

in VSB as shown in the following equation:

Vt = Vt0 + γ((cid:112)2ΦF + VSB −(cid:112)2ΦF ),

(5.5)

where Vt0 is the threshold voltage when VSB = 0, γ is the body effect coefficient, ΦF

is the Fermi-potential and they are all process dependent parameters.

The body effect occurs when VSB (cid:54)= 0 which is the case of all flying NMOS switches in

n-well process. Obviously, body effect increases the transistor threshold voltage, which

consequently increases the Ron, and in extreme cases the switch will not turn on. One

solution to avoid this scenario is to increase the gate signal to maintain the Ron. Other

techniques involving forward body biasing can be also emplyed, but it all depends on

the structure and the technology used.

59

In our design, we use NMOS switch because Ronn < Ronp, and it has lower parasitics

than TG. The charge injection effect is minimized by using higher flying capacitor.

In

addition, gate signal level-shifting is implemented to overcome any possible body effect in

the flying switches.

5.1.2 IC Solar Cells

The on-chip solar cell is basically a photo-diode that generates electricity when it is exposed

to a light source. The light photons excite the electrons in the PN junction which allows

them to move to higher potential. This is translated to an electricity generation. In CMOS

technology, there are two main on-chip PN junctions: D1 and D2 which can be found between

the channel and substrate, see Figure 5.2. Prior works show how these photo-diodes can be

utilized to generate electricity [35–42]. Obviously,D1 provides more energy than D2, because

D1 is exposed and close to the surface. However, if D1 is used, the energy generated by D2

will be lost. In [36], an on-chip solar cell is implemented in 0.5µm CMOS technology, by

utilizing D1 and grounding D2. This solar cell achieves 20µW/mm2 in exposure to 60klux4

light source. On the other hand, a 60µW/mm2 is achieved using D2 under 25klux ultraviolet

light source using 130nm CMOS technology [38]. In this context, It is worth mentioning that

the surface area shape of the on-chip solar cell greatly effects the power density. For example,

a fractal on-chip solar achieves 6% improvement compared to rectangular counterpart [42].

In this work, the solar cell is not considered for same die integration. However, a photo-diode

power density Pph(cid:3) = 20µW/mm2 is selected based on [36], that helps in choosing the power

rating of the converter.

4Lux is the unit of light intensity, the daylight intensity without exposing to direct sun light is 10-25klux.

60

Figure 5.2: Photo-diodes available in CMOS process. Reproduced from [36].

5.1.3 IC Capacitors

The on-chip capacitor is usually associated with parasitic capacitance that is formed between

top-plate, bottom-plate and ground due to the fabrication process, see Figure 5.3 where α

is the ratio of the bottom-plate parasitic capacitor, CBP , to the flying capacitor and β is

the ratio of the top-plate parasitic capacitor, CT P , to the flying capacitor. In fact, α is at

least two times β, because of the minimum insulator diameter of CBP compared to CT P .

For FSC configuration, CT P maintains the charge because it does not have a path to fully

discharge in whole switching period. Therefore, only CBP is considered in this work. The

effect of the parasitic capacitors on the converter efficiency is discussed in section 5.3.

There are many ways to implement capacitors on-chip. Excluding exotic capacitor con-

figurations like trench capacitors, there are two main implementation for on-chip capacitors

CMOS technology as following:

1. Conductor-insulator-conductor capacitor: this kind of capacitors are formed by sand-

wiching an insulator layer between two conductor layers like poly-silicon layer (PIP

capacitor) or metal layers (MIM capacitors). This type of capacitor has lower ca-

pacitance density (C(cid:3)) than MOS capacitors but features less parasitic capacitance.

61

MOM, metal-oxide-metal, is also considered another subset of conductor-insulator-

conductor capacitors, but it requires no additional masks as it exploits the oxide layer

as an insulator. However, MOM has poor capacitance density and substantial parasitic

capacitor.

2. MOS capacitor is implemented by connecting the drain and source terminals of a

transistor together forming one terminal for the capacitor, and the gate works as a

second terminal for the capacitor. This configuration requires no additional masks,

but the capacitance is highly nonlinear and changes based on Vgs value.

For C5 technology, an PIP capacitor is available with a C(cid:3)5 of 950aF/µm2. Therefore,

PIP is used to implement all capacitors on-chip including Cin, Cout and flying capacitors.

Figure 5.3: On-chip PIP capacitor: (a) simple model, (b) 3D layout structure.

5This is a design parameter provided by MOSIS, but after pos-layout extraction using Cadence Diva tool,
only a 790aF/µm2 is achieved. This reduction results in 20% increase in the die area. Hence, the whole
design is reevaluated based on the extracted value.

62

(a)SubstrateoxidePolyPolyCβCαCCαCβCTop-plateBottom-plateFlying capacitorParasitic capacitor(b)5.2

IC FSC challenges

In this section, two main challenges, i.e startup and parasitic loss, for monolithic FSC imple-

mentation are discussed. Solutions to overcome these challenges are proposed. The circuit

implementation with simulation results are to be discussed in the following chapter.

5.2.1 Startup Mechanism

In energy harvesting (EH) circuits like our application example, the startup circuit is very

critical. It provides the system with adequate power for proper operation. The minimum

supply voltage for standard CMOS process is 36mV based on the well-known Meindl limit6

[43]. However, MOS transistor based on EKV model7 requires at least Vgs = 4UT to start

operating in weak inversion where the transistor exhibits high channel resistance, before it

still is not considered as completely turned off [44]. Note that, UT is the thermal voltage

which approximately equals to 26mV at 300 K. Based on this discussion, we consider 104mV

to be the minimum voltage to bias a transistor in normal conditions.

The ambient power sources are low voltage in nature, 10 − 600mV , which complicates

the startup process. Therefore, external startup circuit is commonly employed [45]. Table

5.1 summaries most ambient sources voltage range and power density [46, 47]. Considering

the minimum voltage requirement mentioned earlier, starting up a circuit from lower voltage

requires passive resonating circuits to build up the voltage. For example, in [45], a mechanical

switch with inductor are used to build up the charge to higher level which can be transferred

to startup the converter. On the other hand, when enough voltage is available like the case

6The Meindl limit defines the minimum operation voltage for a binary switching signal transfer.
7EKV defines three operating regions for MOSFET: weak, moderate and strong inversion. The MOSFET

drain current in weak inversion has an exponential relation to the Vgs.

63

Table 5.1: Ambient sources.

0.3-0.6V

Voltage range Power density
10-100µW/cm2
<15mW/cm2
50-300µW/cm3
1.5µW/cm3

25mV

Source

Indoor light

Outdoor light

Photovoltaic (PV)

Piezoelectric (PZ)
Thermoelectric (TEG)a
aReported values based on 1 K temperature difference between cold and heat plates.

vibration

∆Temperature

0.5V

of a solar cell, an active solution can be used to startup the circuit. It can can be realized

using RO with auxiliary charge pumps (AQP) [21, 48–51].

Other possible startup mechanisms are reported in [52]. Most reported startup circuits

employ magnetic elements except for the RO where no inductor is required. Owing to

inductor-less solution, we think RO with AQP is more applicable to our application. RO

circuit with non-overlapping clocks generator is shown in Figure 4.11, the frequency of os-

cillation can be found using (4.25), (5.2), (5.1) and substituting τd = CL(Rp + Rn).

The AQP is usually implemented using Dickson CP or its modified variants; NCP1,

NCP2 and cross-coupled charge pump (CCCP). The improved Dickson CPs are developed

originally to solve the threshold voltage drop problem in conventional Dickson CP. However,

the output stage still suffers from the body effect, which increases the threshold voltage as

we discussed earlier. Therefore, body biasing techniques have been proposed to address this

issue [50, 51, 53]. Nevertheless, the body biasing requires extra circuitry and complicate the

startup circuit. To startup the system in this work, we consider using RO integrated with a

simple 8-module CCCP as shown in Figure 6.3. The detailed implementation of the startup

circuit is to discussed in chapter 6.

64

5.2.2 Parasitic Capacitor

The parasitic capacitance loss is unavoidable in IC design, as shown in Figure 5.7 for a

simulated converter losses. The effect of the parasitic capacitors on the gain, startup and

charge consumption of linear and Fibonacci-like SC converters is investigated in [54, 55],

showing the gain dependency on parasitic capacitors and number of stages. Therefore,

techniques to minimize the parasitic effect would improve the converter performance.

In general, the charge stored in the bottom-plate parasitic capacitor (Qp) due to charging

process either be lost, delivered to the load or recycled. Delivering Qp to the load (intrinsic

Qp recycling) is quite interesting [56]. However, this technique is a structure-dependent and

it is not practical in FSC converters. A novel technique to minimize the parasitic power loss

PBP was proposed in [57] by identifying the flying capacitor with highest bottom-plate loss.

Then, the SN is configured in a way such that this capacitor is discharged once in each three

cycles. This resulted in almost 50% reduction in the the power loss (14.68µW to 7.47µW

in Table III [57]). Nevertheless, this technique can not be implemented in FSC structure

because all flying capacitors in FSC are switched simultaneously in each switching cycle.

Another approach to perform charge recycling is to transfer Qp in the dead-time periods.

As mentioned earlier, SC converters are driven by two non-overlapping clocks. Therefore,

besides the charging and pumping (active) states in SC converters, there are two dead-time

zones to allow Sa to turns off completely before Sb turns on and vice versa, see Figure 5.4(a)

in a two-capacitor FSC example. Obviously, all CBP is charged to Qp during the charging

state and discharged to zero in the pumping state. Hereinafter, by QR is referred to the

Qp dead-time recycling. In [58], QR is proposed using an additional capacitor, CREC , that

works as a charge reservoir. In the first dead-time period, 1

2Qp is stored in CREC , and then

65

the same charge is pumped back to the CBP in the second dead-time interval. Therefore,

effectively 1

2 Qp is conserved in each switching cycle. However, adding CREC increases the

area on-chip, which is always something we need to avoid in IC design.

Figure 5.4: Dead-time charge recycling (QR): (a) Timing signals, (b) FSC converter.

In [59–61], the same concept of QR is employed, but now using internal CBP owing to

the interleaved SC structure. However, this comes in price of adding additional components.

Inspired by this literature, dead-time charge recycling can be used in the proposed FSC

converter by exploiting the availability of CBP in each other state without transforming the

converter into interleaved structures nor adding additional capacitor. Therefore, a charge

recycling unit (QRU) is proposed to transfer and maintain the charge between parasitic

capacitors, CBP1

, CBP2

and CBP3

, during the dead-time periods. Note that, the dead-time

periods should be long enough for Qp to be completely transferred from one capacitor to

another. On the other hand, increasing the dead-time period increases the voltage ripples and

may also affects the conversion performance. Therefore, optimizing the timing is essential

to perform a safe charge recycling. The implementation of QR circuit is to be discussed in

chapter 6.

66

C1SaSqrSbC2CBP2SaSbVoutVinSbCBP1SaSa(b)SaSbSqrTs(a)5.3

IC FSC Converter Analysis

The area on-chip (At) and Ploss are considered the main constrains in IC SC converter

design. The component sizes and optimal switching frequency can be determined by At and

Ploss constrained optimization [27, 62, 63]. In the following, these two factors are studied

and analyzed.

5.3.1 Die Area Analysis

The total area on-chip can be found using the following:

AT = Aph + Ac + Asw + Acnt,

(5.6)

where Aph is the area of the photo-diodes, Ac is the area of all capacitors, Asw is the area of

all power switches and Acnt is the area of the control unit. For sizing and optimization, only

Ac and Asw are considered because they have great effect on the Ploss function as will be

discussed next section. The control unit constitutes negligible area on chip compared others

and hence can be neglected. The photo-diodes occupy the largest portion of the die area.

Nevertheless, optimizing photo-diode is out of the scope of this work. Therefore, Aph can be

estimated using the following equation:

Aph =

P ph
Pph(cid:3)

.

(5.7)

where Pph is the required input power to our circuit. Also, the Ac can be estimated by

dividing the total required capacitance Ct by the capacitance density C(cid:3). Finally, Asw is

the gate width multiplied by the gate length. Then, the total area on-chip in m2 excluding

67

the photo-diodes can be found as following:

At =

Ct
C(cid:3) + LminWt.

(5.8)

One note worth mentioning here, At is mainly determined by total capacitance on-chip,

and having same total capacitance on chip for two SC different topologies does not imply

equal number of stages (check Figure 2 in [63]). However, FSC has the least number of

stages for any VCR compared to other topologies and the capacitance can be optimized.

Consequently less number of switches are required which simplifies the control circuit.

5.3.2 Power Loss Analysis

Let us now analyze the power loss in IC FSC converter. As discussed in section 4.1.2, Ploss

of FSC converter can be quantified using (4.16) and (4.21). However, gate driving loss is

usually considered the main source of Psw in IC design, all other losses can be neglected

because of their minimal effect. Therefore, Psw of IC FSC can be calculated as following:

m(cid:88)

Psw = fs

j=1

(W L)jCoxV 2
gsj

.

(5.9)

where W LCox is the gate capacitance of jth transistor. In addition, the bottom-plate power

loss (PBP ), due to charging and discharging CBP , is very critical and needs to be analyzed

and optimized. PBP can be quantified using the following equation:

k(cid:88)

i=1

αCfi

V 2
BP i

.

PBP = fs

(5.10)

In discrete implementation of FSC, once a switch is selected, the transistor width can not

68

be optimized. On the other hand, the switch is designed from scratch in IC implementation.

In fact, optimizing both the capacitor and switch sizes minimize the Plin losses. Therefore,

both (4.9) and (4.10) are used here. To simplify the analysis, a device resistivity (ρx) with

minimum length Lmin is defined and can be quantified as following:

ρx =

Lmin

µnCox(Vgs − Vtn)

,

(5.11)

where Lmin equals 0.6µm in C5 technology. Based on the foregoing discussion, the total

Ploss of IC FSC can be found using the following:

k(cid:88)

i=1

Ploss = fs

αCfi

V 2
BP i

+ fs

m(cid:88)

j=1

(W L)jCoxV 2
gsj

+ I2

out

(cid:118)(cid:117)(cid:117)(cid:117)(cid:117)(cid:116)

(cid:17)4

(cid:16) k(cid:80)

aci
i=1
C2
t f 2
s

(cid:17)4

,

arj

(cid:16) m(cid:80)

j=1
W 2
t

(5.12)

4ρ2
x

+

In conclusion, the constrained optimization can be used where the Ploss in (5.12) is the

objective function and At in (5.8) is the constrain. The design flow chart for designing an

IC FSC converter is depicted in Figure 5.5.

5.4

IC FSC Converter Design

Using the same converter in chapter 4, this section explores the design and optimization

of the converter specifications. The converter is redrawn in Figure 5.6 including parasitic

capacitors. Table 5.2 shows the C5 parameters used to quantify At and Ploss functions. Note

that, PIP capacitance density is found to be 790aF/µm2. In this work, we are limited to

69

Figure 5.5: IC FSC design procedure flow chart.

70

Specifications: Vin, Vout,  Vout, Pout, ηUsing VCR defined in step 1  Apply FSC synthesis procedure in Chapter 3Define ar, ac, av vectors of the worst case using procedure in Chapter 4Using any constrained optimization method like GA, for specified Pout find global minima of Plosswith At as a constrainIs η acceptable?Final Circuit realization with components sizesNoYesChoose technology and define all parameters (µn, Cox, Lmin, Vtn, α) and choose capacitor types.Derive: At based on (5.6)and Ploss (Iout, Vout, Ct, Wt, fs, Vgs) based on (5.10)Table 5.2: C5 technology parameters

Parameters

NMOS

PMOS

µ

(cm2/V.s)
445

201

tox
(nm)

13.5

13.5

Cox

(nF/cm2)
260.4

260.4

Vt
(V )

0.7

-0.9

Lmin
(µm)

0.6

0.6

4mm2 die area8. However, considering the IO pads9, which usually constitutes close to 30%

of the area, only 2.8mm2 is assumed to be the available active area. In addition, the input

and output capacitors are integrated on-chip and sized as the minimum flying capacitor.

Therefore, we can define the total area (At) constrain as following:

At = 0.002Ct + 0.6Wt ≤ 2.8mm2,

(5.13)

where the units of Ct and Wt are in pF and µm, respectively. Second step is to find

optimized parameters using charge vectors found in section 4.2.1. They are rewritten here

for convenience:

(cid:20)

(cid:21)

(cid:20)

(cid:21)

ac

=

2 1 1

(cid:20)

(cid:21)

3 2 2 1 1 1 1 1 1 1 0 1 0 1 0

(cid:20)

(cid:21)

ar

=

Therefore, C1 = Ct/2, C2 = C3 = Ct/4 using (4.11). To account for zero-elements in ar,

we assume the worst case for switches Sg1, Sg3, Sg5 as 1, 1, and 1/2, respectively. They are

found using the whole ar matrix in section 4.2.1. Then, the switch-on resistances can be found

using (4.12) as following, W1 = 6Wt/37, W2 = W3 = 4Wt/37, W4−10 = Wg1−g4 = 2Wt/37

8This project is considered to be fabricated under MOSIS Educational Program (MEP), where the die

area is limited to 4mm2.

9IO pads connect the IC to the outside world, and it is extremely important for proper operation. Some

details for the used IO pads are available in chapter 6.

71

Figure 5.6: Proposed FSC with parasitic capacitors.

and Wg5 = Wt/37, where Wt is the total width for all switches. Now, let us find the device

resistivity, ρx using (5.11). First, consider the worst case where Vs = 3Vin which is the case
of Sout, Vgs − Vtn = Vg − 3Vin − Vtn. Using Vg ≥ 2, Vin = 0.3 and Vtn = 0.7, ρx is found
to be 0.0129Ω · m. Combining these results, the optimal Req can be found using (4.9) and

(4.9), as following:

(cid:115)

Req =

256
C2
t f 2
s

+

79
W 2
t

,

(5.14)

Let us now find the power loss of the bottom-plate parasitic capacitor. By analyzing the

voltage across CBP in each phase, it was found that VBP 1
relating Vin = 1

5 Vout, PBP (Ct, fs, Vout) can be found as following:

= VBP 2

= 1

2 VBP 3

= Vin. By

PBP = 0.09αCtfsV 2

out,

(5.15)

Then, by using Vgs = 2 and subsisting (5.9), (5.14) and (5.15) in (5.12), the total Ploss

72

C1S1CoutVoutVinSg2CBP1CTP1S2S3C2S4CBP2CTP2S5S6Sg1Sg4C3S7CBP3CTP3S8S9Sg3Sg5Soutof the designed FSC converter can be found as following:

Ploss = 0.09αCtfsV 2

out + 6.25 × 10−9Wtfs + I2

out

(cid:115)

256
C2
t f 2
s

+

79
W 2
t

.

(5.16)

Referring to the design flow chart in Figure (5.5), we are ready for optimization step. We

formulate our optimization problem using objective function as Ploss in (5.16), the constrain

as At found in (5.13), and using three variables Ct, fs and Wt. Multiple runs have been

conducted using GA for 10µW output power and the parameters are selected as following:

960pF total capacitance, 200kHz switching frequency and total transistor width of 388.5µ

m. The total power loss is found to be almost 6.85µW with 59% efficiency. Figure 5.7 shows

the contribution of losses. It can be seen that PBP loss constitutes large amount of the total

power loss. Applying the dead-time charge recycling (QR) decreases the Ploss by almost

27% as shown in Figure 5.7, which is translated to an increase of 15% in the total efficiency.

More details about the advantages gained from applying QR is discussed in chapter 6.

Figure 5.7: Power losses shares with and without charge recycling (QR).

73

27% reduction5.5 Summary

In this chapter, challenges of implementing IC FSC converter are discussed. Devices imple-

mentation is covered. The design procedure for IC FSC converter is illustrated through the

solar energy harvesting case. It is shown that parasitic effects have a substantial effect on

the converter performance. Charge recycling (QR) technique is implemented to minimize

the PBP loss and improve the converter overall efficiency. Ideally, this technique reduces the

PBP by almost 50% which is translated to a 27% reduction in Ploss, and 15% improvement

in the converter efficiency. In next chapter, the whole system design, optimization, layout

and post-layout simulation are to be covered.

74

Chapter 6

Analog and Digital Circuit Design

Following previous chapter, this chapter continues to discuss the details of designing IC FSC

and concludes with the post-layout simulation results. The complete IC circuit is shown

in Figure 6.1. The startup circuit includes ring oscillator, auxiliary cross-coupled charge

pump (CCCP), current starved ring oscillator (CSRO) and non-overlapping clock generator

(N-OV). This circuits is designed to generate clocking signals and DC voltage of >1V to

operate the controller.

Figure 6.1: FSC IC building blocks.

75

ROCSROCCCPN-OVGearboxcontrolLSQRUSa, Sbclka,bφcompvariable FSC converterPVcellSg1-Sg5VoutGearboxcontrolQRULSN-OVCCCPROCSROVinon-chipto CBP caps.VcpThe gearbox controller can be implemented in two possible structures as shown in Figure

6.2, either using a feed-forward scheme which is the same circuit used in chapter 4, or in a

feedback topology if there is a way to sustain a reference voltage. For the latter, a hysteresis

controller can be employed. It senses the output voltage and sends count up/down signal to

a counter which drives the gearbox switches to sustain the load voltage between VH and VL.

The feedback controller provides better stability and controllability over VCR. However, the

feed-forward controller is used here due to the lacking of the high reference voltage. In the

following sections, proposed startup, clock generation and comparator circuits are discussed.

Then, the simulation results are presented with discussion.

Figure 6.2: Possible FSC controller structures: (a) Feed-forward, (b) Hysteresis controller.

6.1 Cross-Coupled Charge Pump Design

Due to its structure, CCCP can boost sub-threshold input voltage to higher levels, but it

suffers from poor efficiency. Therefore, CCCP is employed as an axillary charge pump for

the system. This circuit requires also two complementary clocking signals. A simple three-

76

R1R2R3R4VDDSg1Sg2Sg3Sg4Sg5ABCVHCounterVLDownUpFSC converterSg1-5InputClockgenerationSa,bhysteresisL  o  a  dVin(b)(a)stage ring oscillator (RO) is used to generate these signals. The complete circuit is shown

in Figure 6.3. Eight-module CCCP is used to boost the voltage to relatively usable level.

Figure 6.3: Three-stage RO with eight CCCP modules.

In C5 technology, p-substrate is used which means NMOS transistors suffer from body

effect when their source and bulk (body) are not shorted. Whereas, the body effect can be

avoided in PMOS transistors by applying PMOS in separate n-wells. Due to the body effect

in NMOS transistors, W/L needs to be optimized achieving minimum channel resistance.

Assuming NMOS transistor is in weak inversion (WI) region, the following derivation can

be applied:

(cid:32)

Id0

e

Ids(W I) =

W
L

Rds(W I) =

(cid:33)

,

−Vds
UT

1 − e

Vgs−Vth
nUT

(cid:33)(cid:32)
(cid:34) ∂Ids(W I)
(cid:17)

∂Vds

(cid:35)−1
(cid:16) nVds−Vgs+Vth

,

e

nUT

(cid:16) LminUT

Id0

Rds(W I) =

1
W

(cid:17)

,

(6.1)

(6.2)

(6.3)

where Id0

is the drain current that flows when Vgs=Vtn, and n is the inversion slope factor

which is sensitive to body effect.

77

2.4pFVinCcpCcpCcpCcpCcpCcp1pFφaφbVcpMn1Mn2Mp2Mp2CCCP1CCCP2CCCP8The MOS channel length is chosen to be the minimum, Lmin. So, one can combine the

explicit first order process parameters,

. Due to the complexity of this deriva-

(cid:16) LminUT

(cid:17)

Id0

tion, only the MOS channel width, W, is considered for optimization alongside with the

flying capacitor, Ccp. Therefore, extensive runs of spice simulation in Virtuoso ADE reveal

that W=6µm and Ccp=240f F has the best performance, and hence they are selected. The

capacitance seen at φa and φb nodes is around 2.4pF and hence a capacitance of the same

value is added to the RO third terminal to ensure smooth oscillation. Also, 1pF capacitance

is added at the CCCP output to minimize the output ripple. The final layout is shown in

Figure 6.4.

Figure 6.4: Layout of RO with CCCP.

The post-layout simulation is shown in Figure 6.5. The unloaded output voltage, Vcp, is

around 2.3V in response to 0.5V input. But, when the CCCP is loaded, the output voltage

drops down. In ADE, a noise that is required for oscillation can be injected in some nodes

by setting initial conditions to zero.

78

2.4pFROCCCP1pFFigure 6.5: Post-Layout simulation for the startup circuit using 0.5V input voltage.

6.2 Clock Generation

To drive the FSC converter, two non-overlapped clocks need to be generated on-chip. Here, a

current-starved ring oscillator (CSRO) is used to generate 200kHz clock signal, which is then

fed to a non-overlapping clocks generator to induce a dead-time period between clka and

clkb. The later two signals are fed to a level-shifter to get Sa and Sb. The circuit realization

is shown in Figure 6.6.

The CSRO is a voltage controlled oscillator. It works like a regular RO discussed previ-

ously in chapter 5, but with additional feature of being controllable through current mirrors,

M1-M4, by controlling the amount of current ID flowing to the inverters. Due to process and

temperature variations in IC fabrication, CSRO is employed here to control and calibrate

the frequency of oscillation to be around the designed value of 200kHz. The frequency of

oscillation equation is derived in [64], and can be quantified as following:

fosc =

ID

nroCtotVDD

,

79

(6.4)

Zoom inwhere Ctot is the total capacitance seen at the output of each inverter which equals to

Cin+Cout+Cro, and nro is the number of inverters.

Figure 6.6: Three-stage CSRO with non-overlapping clock generator.

After optimization process of considering both area and power consumption, three-stages

CSRO have been selected with current mirrors sized at W =5µm and L=1µm and load

capacitor of 375f F for each inverter. The frequency of oscillation is calibrated off-chip using

Vb pin. Also, note that the dynamic comparator requires undistorted clock signal, so the

output of CSRO is buffered by an inverter. For the non-overlapping circuit, 100f F and

off-chip delay capacitors, Cda and Cdb are added to control the dead-time off-chip.

Level-shifter is also employed for another voltage boosting stage to overcome the need

of higher gate voltage in the high nodes like the source of Sout in the main FSC converter.

The level-shifter circuit is shown in Figure 6.7, MLS2 turns on when clka goes high pulling

node x to zero. Then, the PMOS transistor, MLS3, turns on and hence Sa node clamps

to VDD2. For next cycle when clkaN is high, MLS1 conducts and it pulls Sa node down

to zero. The same procedure are repeated for Sb level shifter. The level-shifter transistor

80

fosc375fFVb375fF375fFVcpM1M2M3M4MPMNIDCdaφcompCSROclkaclkb0.1pF0.1pFCbdsizes are optimized for the delay between clk signals and driving signals, Sa and Sb, and the

power consumption. The sizes Mn=10.5µm/0.6µm and Mp=1.5µm/0.6µm are selected.

Figure 6.7: Level shifters with QRU.

Figure 6.8: Clocks generation layout.

Moreover, a charge recycling unit (QRU) is used to transfer the charge between parasitic

capacitors. QRU consists of a simple AND gate with minimum sized transistors Mqr1 and

Mqr2 with a size of 1.5µm/0.6µm. The AND gate output, Sqr goes high in the dead-time

where Sa and Sb are low. Sqr is driving the charge-transfer transistors, Mqr1 and Mqr2.

81

SqrCBP1CBP2CBP3Mqr1Mqr2QRUVDD2clkaNMLS3MLS4clka10.50.6mm10.50.6mm1.50.6mm1.50.6mmMLS1MLS2SaVDD2clkbNMLS7MLS8clkb10.50.6mm10.50.6mm1.50.6mm1.50.6mmMLS5MLS6Sb1.50.6mm1.50.6mmxyCSRON-OVQRULevelshifterTherefore, during the dead-time, the charge accumulated in the parasitic capacitor of C2 is

transfered to the parasitic capacitors of C1 and C3 fulfilling the charge recycling concept.

The post-layout simulation for this chain of clock generation is conducted using Virtuoso

ADE in Spectre simulator and is shown in Figure 6.9. The frequency of oscillation is around

200kHz and Sqr is generated as expected.

Figure 6.9: Clocks waveforms (post-layout simulation).

6.3 Comparator

The comparator circuit is considered the bottle neck of achieving ultra-low power consump-

tion for the control circuit [64]. There are two main comparator circuit operation schemes,

static and dynamic as shown in Figure 6.10. For both, the comparator consists of preampli-

fier stage, latch circuit and sometimes separate output stage. The preamplifier stage, which

is usually implemented by a transistor differential pair, amplifies the input difference voltage

to make it readable by the latch circuit. The latter decides whether the vp is greater or less

than vn and outputs a logic signal, VDD or ground, and sometime referred to as a decision

circuit.

82

5µsFigure 6.10: Comparator functional blocks: (a) Static comparator. (b) Dynamic comparator.

Generally, the static comparator consumes more power than the dynamic but requires

no clocking signals. Figure 6.11 shows the CMOS implementations for a static and a dy-

namic comparator. It can be noticed that static comparator can be implemented using a

conventional operational amplifier (opamp) CMOS circuit without internal compensation or

feedback. The dynamic comparator requires a clock signal which is connected to the pre-

amplifier stage as well as the latch circuit, M8-M11. Dynamic comparators are preferred

due its low power profile [21]. Both comparators are studied and simulated for comparison

purposes. Dynamic comparator is selected due its highly power efficient performance.

Figure 6.11: CMOS comparator structures.

83

LatchAA(vp-vn)vpvnclkPre-ampliferLatchAvpvnA(vp-vn)(a)(b)VoutVoutVoutvnvpM1M2M3M6M5M7M8M4RVoutclkvnvpM1M2M3M4M5M6M9M8M10M11M7M12M13Static comparatorDynamic comparatorFor designing comparators, there are many static and dynamic merits need to be con-

sidered [65]. However, for this application we consider two parameters, offset voltage (Vos)

and power consumption. The offset voltage is the minimum amount of input voltage differ-

ence that makes the output reaches mid-point. This is can be characterized easily for static

comparators by applying DC simulation. On the other hand, finding Vos for the dynamic

comparators is not trivial because transient analysis is needed due to the presence of a clock-

ing signal [66, 67]. Nevertheless, a simple technique to estimate the offset is developed in

three steps:

(1) Vos of a static comparator is found using DC simulation as mentioned earlier.

(2) The same static comparator is simulated in transient analysis, and we measure the

minimum (vp-vn) that makes the output reaches mid-point, which is defined as Vostr .

Then, a fitting factor is defined as x=Vos/Vostr .

(3) Using the same transient simulation, the dynamic comparator offset voltage can be

approximated as x ∗ Vostr , where Vostr here is the dynamic comparator offset voltage

in transient response.

The two comparators are simulated using Cadence Spectre simulator in C5 technology.

For both, a 0.5V reference voltage (Vref ) is connected to vn terminal and an input voltage

that varies from 0-1V is connected to vp, it is expected that when V in > Vref the comparator

outputs logic 1, i.e Vout=VDD, and it outputs logic 0, i.e Vout=0, otherwise. The simulation

for the static comparator with R=100kΩ shows that it has an offset voltage of 62mV, and

consumes 1.6µW. On the other hand, the dynamic comparator consumes only 51nW which is

almost thirty times less than the static comparator, and it experiences a 74mV offset voltage.

Both comparators are loaded with 5pF capacitor. The dynamic comparator is operated using

84

200kHz which is provided by the CSRO circuit as discussed earlier, this is enough because

input fluctuation is much less frequent [64]. Although we ignore the system propagation

delay when designing the comparator, this parameter can be detrimental to the operation of

the controller when there is a fast and abrupt variation in the input signal, i.e Vin changes

from 0.3V to 0.5V in > φcomp rate. This results in violating Nyquist rate theorem and hence

unwanted results can be reported. The transistors in both circuits are sized to the inverter,

i.e Wp=3µm and Wn=1.5µm and Lp=Ln=0.6µm. However, the differential pair use wider

transistors, M1 and M2 with W=7.5µm, to provide enough gain for amplification.

Figure 6.12: Proposed CMOS dynamic comparator.

The dynamic comparator is chosen for the intended circuit. However, this comparator

makes a decision each clock which means it goes to zero as a stable state after clamping

the output to VDD in the case of higher positive terminal. This needs to be overcome by

a mono-stable latch circuit, so the output will remain at the decision level, 0 or 1, until

85

VoutφcompvnvpM1M2M3M4M5M6M9M8M10M11M7ABYclkDQpQnQpQnDclkN-T DFFN-T DFFNANDTable 6.1: Truth table for N-T DFF.

Input
Output
D φcomp Qp Qn
↓
0
↓
1

0
1

1
0

it makes another transition based on the change of the input. A Negative-Triggered D-flip

flop (N-T DFF) is used to clamp the output voltage to either VDD or gnd. The truth table

of N-T DFF is shown in Table 6.1. The proposed dynamic comparator with N-T DFF is

shown in Figure 6.12. The layout is shown in Figure 6.13, and it occupies an area of only

0.003mm2. The post-layout simulation is performed to account for parasitics and to allow

re-optimizing the transistors sizes. Figure 6.14 shows the waveforms for Vout, Vref and Vin.

It can be seen that the proposed comparator outputs VDD when Vin > Vref which meets

the expectations. The gearbox controller layout is depicted in Figure 6.15.

Figure 6.13: CMOS dynamic comparator layout.

86

Figure 6.14: Post-layout simulation for the proposed dynamic comparator.

Figure 6.15: The gearbox controller layout.

87

00.10.20.30.4Time(ms)00.51Voltage(V)VoutVrefVinLogicCOMP2COMP3COMP1Resistors6.4

IO Pad Design

The IO pads design is very critical as it connects the IC to the outside world.

IO pads

should have Electro-Static-Discharge (ESD) to protect the IC from excessive charge that may

damages the devices, especially switches. There are very advanced techniques to implement

ESD. However, in the proposed IC, a simple ESD circuit is used as shown in Fig. 6.16. Two

diodes, D1 and D2, with large areas are used to clamp the internal pin to the power rails,

VDD or gnd. When the pin is exposed to a high static voltage greater than VDD+Vdiode,

D1 conducts discharging the excessive charge and clamping the pin to VDD as illustrated by

the red line in Figure 6.16. On the other hand, if the IO pin experienced a negative charge

with voltage less than the diode voltage, D2 conducts discharging it to gnd and grounding

the pin potential (the green dashed line).

Figure 6.16: ESD for IO pads.

This circuit provides a relatively good ESD protection and consumes no active switch-

ing or complex structure. However, for circuits that are meant for commercial use, other

complicated ESD can be applied.

88

VDDD1D2IO Padinternal pinFigure 6.17: Layout of an IO pad with ESD.

6.5 Final Layout and Post-Layout Results

The final chip layout including flying capacitors, input and output coupling capacitors, is

shown in Figure 6.18, and it occupies 4mm2 die area. Also, it can be seen that the controller

circuit with the converter switches occupy only 0.0754mm2 which is almost 1.89% of the

total die area. For the pin configurations, the main IC pins are Vin, Vout, gnd and the

tuning pins (Cda, Cdb and Vb). The biasing pins (VDD1, VDD2 and VDDRef ) are added as a

backup if the on-chip auxiliary charge pump fails to provide power for the circuit. The gate

signal pins (Sqr, SaRO, SbRO, Sg1−g5) are utilized for debugging purposes. All other pins

are used for additional FSC cells. Table 6.2 defines all IO pins.

Before tape-out, the design-rule-check (DRC), the layout-vs-schematic check (LVS) and

parasitic extraction have been done using Diva tool in Cadence. Analog Design Environment

(ADE) has been used to perform all simulations using Specter simulator. All circuits have

been re-optimized after extensive round of simulations including the parasitics, referred to

as a post-layout simulation.

89

PadD1D2Table 6.2: IC pin configurations.

Description
Mid-point connection for cell4
Bottom-plate flying capacitor connection for cell4
Top-plate flying capacitor connection for cell4
Input for Sout switch
Output for Sout switch
Gate driving signal for Sout switch
Probing gearbox control signal for Sg5 switch
Probing gearbox control signal for Sg4 switch
Probing gearbox control signal for Sg3 switch
Probing gearbox control signal for Sg2 switch
Probing gearbox control signal for Sg1 switch
Probing on-chip ring oscillator signal {S b}
Probing on-chip ring oscillator signal {S a}
Probing on-chip QRU signal
Dead-time tuning capacitor for on-chip N-OV circuit
Dead-time tuning capacitor for on-chip N-OV circuit
Oscillation frequency tuning for on-chip CSRO
Input for on-chip auxiliary CCCP
Input for the configurable FSC chip
Ground reference for the configurable FSC chip
Output of the configurable FSC chip
Output of the on-chip auxiliary CCCP
Backup biasing source for all controller circuits
Biasing source for level shifters

No. Pin
1
Tk4
2
TB4
3
TP4
4
SOin
5
SOout
6
SOg
7
Sg5
8
Sg4
9
Sg3
10
Sg2
11
Sg1
12
SbRO
13
SaRO
14
Sqr
15
Cda
16
Cdb
17
Vb
18
VcpIn
19
Vin
gnd
20
21
Vout
22
Vcp
23
VDD1
24
VDD2
25
VDDRef Backup biasing source for gearbox reference voltage
26
Sa
27
Sb
28
Tin1
29
Tk1
30
TB1
31
TP1
32
Tin2
33
Tk2
34
TB2
35
TP2
36
Tin3
37
Tk3
38
TB3
39
TP3
40
Tin4

External gate driving signal to all additional FSC cells
External gate driving signal to all additional FSC cells
Input connection for cell1
Mid-point connection for cell1
Bottom-plate flying capacitor connection for cell1
Top-plate flying capacitor connection for cell1
Input connection for cell2
Mid-point connection for cell2
Bottom-plate flying capacitor connection for cell2
Top-plate flying capacitor connection for cell2
Input connection for cell3
Mid-point connection for cell3
Bottom-plate flying capacitor connection for cell3
Top-plate flying capacitor connection for cell3
Input connection for cell4

90

Figure 6.18: The chip layout with pin configuration.

The final IC has been simulated for 0.3-0.6V input voltage range and various load changes.

Previously in the design stage, the optimal efficiency was found at 59% but with assumption

of regulated output voltage of 1.5V. The adaptive VCR is built using forward controller, and

hence the system does not respond to load change. As a result, the output voltage is not

fully regulated, see Figure 6.19 that shows the transient response of the output voltage for

two load cases, no load and 1µA load. With only a capacitive load, the system provides an

output voltage of 1.4V and it drops to 1.24V with 4.6mV ripple in response to 1µA load.

This µW power level is useful in many applications. For example, a neural signal detector

presented in [68] consumes only 0.78µW. The miniaturized pacemaker proposed in [69],

requires 1.2µW using 1.3V supply voltage.

In addition, a wireless temperature/pressure

sensor is presented in [70] for continuous intraocular pressure monitoring, and dissipates

91

2.3µW with 1.5V source. Considering WSN application,we can point to two examples. The

first one is a wireless receiver that works on a wake-up approach, and consumes around

2.4µW at 1V supply [71]. The second example is a CMOS image senor found in [72], which

can be used for monitoring, and it requires 1.36µW power. These are few examples of many

for µW and sub-µW level applications.

The efficiency profile for different loads is shown in Figure 6.20, including the theoretical

case where no charge recycling is applied. Only an error of less than 6.3% has been noticed

in the efficiency between the post-layout simulation and the analytical estimation of Ploss

in equation (5.12). The converter has a peak efficiency of 47.5% for 5.2µA load current,
and maintains efficiency ≥ 40% for 2.4∼8.8µA. Nevertheless, the output voltage is not fully

regulated to the load due to the absence of a feedback controller. Figure 6.21 shows the
output voltage verses different loads. It is found that Vout ≥1V is available for ≤ 6.5µA load
current. Regarding the VCR, a discrepancy of ≤ 6% is noticed between the expected and
simulated VCR as depicted in Figure 6.22 for Vin ≤ 520mV, and it is kept under 11.5% for

0.52-0.6V input voltage.

Finally, a performance comparison between the proposed converter and other state-of-

the-art converters is summarized in Table 6.3. One can notice that the proposed controller

exhibits low power consumption which is almost 2.8x less than the minimum controller, i.e

342nW compared to 971nW in [49]. Also, the proposed converter provides a monolithic

solution, i.e no off-chip components are required, as same as [73]. But converters in [73]

and [74] use a native NMOS device which is not available in some process like C5. Therefore,

they are considered as unportable1 compared to all others. In terms of efficiency, despite

1Technology portability means that the proposed structure can be implemented in all standard technolo-

gies because there are no special devices involved in the design.

92

the fact that [49] and [75] score higher than 80%, a high input voltage was used in [49],

1-1.5V, along with dual MIM capacitor which has lower parasitics and it is not available in

C5. For example, the parasitic ration, α, in [75] is only 1%. Moreover, the losses reported

in [75] are around 31µW, for all losses, which means for few µW the efficiency will drop

below 30%. The efficiency of the proposed converter is relatively below 50% because we are

limited to PIP capacitors which has almost 13% parasitic ratio and starting up from low

voltage (300-600mV).

Figure 6.19: Efficiency verses Iout for the proposed FSC converter with 300mV input voltage.

6.6 Summary

In this chapter, the details of designing, simulating and testing the IC FSC has been covered.

The proposed converter boosts the low input voltage to 1.4V (with no load) and 1.24V (with

µA load) which can be used to power up a battery-less system like a wireless sensor node

and/or an implanted biomedical sensor which usually requires few µW to wake-up and sends

93

Zoom inripple ≤ 4.6mVVout,avg =1.24VFigure 6.20: Efficiency verses Iout for the proposed FSC converter with 300mV input voltage.

Figure 6.21: Vout verses Iout for the proposed FSC converter with 300mV input voltage.

94

051015Iout (A)01020304050 (%)w/o QR (analytical)with QR (analytical)Post-layout simulation10-410-21100Iout (A)00.511.5Vout (V)AnalyticalSimulationFigure 6.22: VCR verses Vin for the proposed FSC converter.

Table 6.3: Performance Comparison with SC converters in the literature.

Reference

JSSC’16 [76] JSSC’15 [49] TCAS’11 [73] TBCAS’17 [74] TCAS’15 [75]a This Worka

Tech. node

180nm

Topology

Dickson

130nm
2-stage
SC doubler

65nm
4-stage
Dickson

Input voltage

0.25-1.1V

1.1-1.5V

0.35-0.48V

Output voltage

1.8-2V

Controller loss

1.18µW

3.3V
971nWb

Capacitor type MIM

Dual MIM

1.4V

3µW

NA

58%

Peak efficiency

48%
70µWc

86.4%

21µW

Yes

Max. Pout
Tech. portable? Yes
aPost-layout simulation
bCalculated from the results
cThree power domains were reported. Low (< 40µA load) , medium (40 − 210µA load) and high (0.21-1mA load)

Restricted

Restricted

645µW

11µW

120µW

Yes

250nm
8-stage
Dickson

0.22V

1.8V

NA

NA

68%

250nm
4-stage
Series-Parallel

0.5- 2.5V

1-1.4V

6µW

MIM

83%

500nm
3-stage
Fibonacci

0.3-0.6V

1-1.4V

342nW

PIP

47.5%

6µW

Yes

95

0.350.40.450.50.550.6Vin (V)2345VCR05101520Error (%)AnalyticalSimulationErroror process information.

96

Chapter 7

Conclusions and Future Work

7.1 Conclusions

In this dissertation, we introduced the variable FSC converter as a good candidate for low

power applications. We first studied the fundamental theory of SC converters where we

proposed a generalized equation to overcome some limitations found in the theory. This

generalization allows to identify the true voltage conversion performance limits of specific

number of k capacitors. In addition, the theory developed in chapter 2 leads us to propose

a unified algorithm to solve the FSC converter synthesize problem as discussed in chapter 3.

Moreover, the power loss analysis and optimization techniques for both discrete and inte-

grated implementation of variable FSC have been discussed throughly in details in chapters

4and 5. Design procedure flow for both converters was proposed. An on-line variable FSC

synthesizer using proposed algorithm in chapter 3 is developed and available as an open-

source tool to help synthesizing FSC converters.

Also in this work, the challenges of implementing monolithic FSC converters were studied,

such as startup and parasitic capacitors losses, see chapters 5 and 6. We proposed a startup

circuit using cross-coupled charge pump with ring oscillator which is a magnetic-free startup.

To overcome the parasitic loss, the dead-time charge recycling was employed by maintaining

the charge in the parasitic capacitors during the dead-time bands. For the controller circuit,

a dynamic comparator was designed and simulated. Despite that fact that this comparator

97

was optimized for the proposed FSC controller, it can be employed in many low power

applications.

7.2 Future Work

In academia, there is always a room for improvement. The SC converter research has a

potential for many applications involving any kind of power conversion. Therefore, for a

future work, the following directions can be considered:

• Design tool: as part of the FSC synthesizer development, the design process of finding

the multiplier vectors, i.e ar, ac and av discussed in chapter 4, can be automated.

• MIMO SC converters:

these converters can be useful in applications that has

many inputs and/or delivering power to different loads with distinct voltage levels.

An overview study has been conducted in [77], and a recent MIMO SC converter is

proposed in [78] where the parasitic capacitors were employed to perform the charge

transfer. Speaking of applications, see Figure 7.1, MIMO SC converter shows a poten-

tial to be utilized as an on-chip power management unit that responds to different volt-

age domains required by on-chip processors [79–81]. In addition, MIMO SC converters

can be employed as a centralized DC-DC converter block in energy harvesting system

extracting energy from more than one ambient source with low voltage profile. There-

fore, the following questions can be further explored to promote the implementation

of these converters: (Q1) How to synthesize MIMO SC converters? Initial thoughts

were proposed in our published work [12], and (Q2) How to model MIMO structure?

which helps understanding the system behavior in response to presence/absence of

some inputs and outputs.

98

Figure 7.1: MIMO SC conceptual blocks for possible applications: (left) on-chip power
management unit for microprocessors, and (right) energy harvesting system.

• On-chip devices: in this work, the on-chip capacitors is considered a major obsta-

cle towards high efficient SC converters. In fact, there are many integarble capacitor

structures like trench capacitor with higher capacitance density. Nevertheless, the ca-

pacitance density can be further increased to 100-200x using on-chip supercapacitors

which can be considered as a game changer device. Supercapacitors were introduced in

1950, whereas the on-chip version took almost 60 years to be demonstrated. These ca-

pacitors are based on double layers structure, and hence they can soak the energy much

faster and deliver them with a relatively constant voltage profile. So, supercapacitor

are featured with higher energy storage, low ESR which enables fast charging and high

surge current delivery. In fact, utilizing on-chip supercapacitors minimize the overhead

die area improving both, the power density and converter efficiency. The on-chip su-

percapacitors are now achievable on silicon substrate as demonstrated in [82, 83], but

it needs some time to be introduced in the normal foundry process. Meanwhile, devel-

oping on-chip supercapacitors featuring low parasitics and using a simple fabrication

process and structure, leads to high efficient on-chip converters.

99

BattPVTEGPZCentralized MISO SCLoadGlobal VDDCentralized SIMO SCVL2VLnVL1• High Gain Operation: Fibonacci converter provides highest gain possible with least

number of capacitors and switches as discussed in chapter 2, which lends itself to a

high gain requirement on-chip. Examples of on-chip high gain systems include LCD

driver [84], LED driver [85] and MEMS, Micro-Electro-Mechanical-Systems, driver

[86], which usually require high voltage operation, i.e >20V. Also, a high VCR is

needed for energy harvesting systems that employ TEG sensors, because the open-

output-voltage of the TEG can be in the range of few tens of mV for 1-2K. So, FSC

issues in high-gain operation can be also investigated to find out the limits and design

requirements.

• AC operation: although in this work we only consider DC-DC operation, the SC

converters can be synthesized in a way to introduce inversion for the Vin or Vout as

verified in chapter 2. Consequently, AC-DC or DC-AC operation can be achieved.

However, high voltage seen at the AC mains (120-230VAC) is considered a major

limiting factor for the fabrication process because most standard technology processes

offer low voltage ratings, e.g C5 has 5V rating. As an example for possible on-chip

inverter implementation, [87] demonstrated a monolithic 120/230VAC to 3.3V inverter

that delivers power to 3mW load targeting Internet-of-Things (IoT) and smart home

applications. But, this on-chip inverter was implemented using two stages, rectification

and step-down dc-dc conversion. Other possible option is to employ only one stage of

SC converter that can performs both operations at once.

100

BIBLIOGRAPHY

101

BIBLIOGRAPHY

[1] F. Ueno, T. Inoue, I. Oota, and I. Harada, “Emergency power supply for small computer
systems,” in 1991., IEEE International Sympoisum on Circuits and Systems, Jun 1991,
pp. 1065–1068 vol.2.

[2] M. S. Makowski and D. Maksimovic, “Performance limits of switched-capacitor dc-dc
converters,” in Power Electronics Specialists Conference, 1995. PESC ’95 Record., 26th
Annual IEEE, vol. 2, Jun 1995, pp. 1215–1221 vol.2.

[3] J. F. Dickson, “On-chip high-voltage generation in mnos integrated circuits using an
improved voltage multiplier technique,” IEEE Journal of Solid-State Circuits, vol. 11,
no. 3, pp. 374–378, Jun 1976.

[4] D. Kilani, B. Mohammad, H. Saleh, and M. Ismail, “Ldo regulator versus switched
inductor dc-dc converter,” in 2014 21st IEEE International Conference on Electronics,
Circuits and Systems (ICECS), Dec 2014, pp. 638–641.

[5] Y. Pascal and G. Pillonnet, “Efficiency comparison of inductor-, capacitor-, and
resonant-based converters fully integrated in cmos technology,” IEEE Journal on
Emerging and Selected Topics in Circuits and Systems, vol. 5, no. 3, pp. 421–429, Sept
2015.

[6] R. W. Erickson and D. Maksimovic, Fundamentals of power electronics.

Springer

Science & Business Media, 2007.

[7] G. Palumbo and D. Pappalardo, “Charge pump circuits: An overview on design strate-
gies and topologies,” IEEE Circuits and Systems Magazine, vol. 10, no. 1, pp. 31–45,
First Quarter 2010.

[8] J. D. Cockcroft and E. T. S. Walton, “Experiments with high velocity positive
ions,” Proceedings of the Royal Society of London. Series A, Containing Papers of a
Mathematical and Physical Character, vol. 129, no. 811, pp. 477–489, 1930. [Online].
Available: http://www.jstor.org/stable/95496

[9] J. A. Starzyk, Y.-W. Jan, and F. Qiu, “A dc-dc charge pump design based on voltage
doublers,” IEEE Transactions on Circuits and Systems I: Fundamental Theory and
Applications, vol. 48, no. 3, pp. 350–359, Mar 2001.

102

[10] A. Kushnerov, “Multiphase fibonacci switched capacitor converters,” IEEE Journal of
Emerging and Selected Topics in Power Electronics, vol. 2, no. 3, pp. 460–465, Sept
2014.

[11] M. S. Makowski, “On performance limits of switched-capacitor multi-phase charge pump
circuits. remarks on papers of starzyk et al.” in 2008 International Conference on Signals
and Electronic Systems, Sept 2008, pp. 309–312.

[12] Y. Mahnashi and F. Z. Peng, “Generalization of the fundamental limit theory in a
switched-capacitor converter,” IEEE Transactions on Power Electronics, vol. 32, no. 9,
pp. 6673–6676, Sept 2017.

[13] S. R. Sanders, E. Alon, H. P. Le, M. D. Seeman, M. John, and V. W. Ng, “The
road to fully integrated dc-dc conversion via the switched-capacitor approach,” IEEE
Transactions on Power Electronics, vol. 28, no. 9, pp. 4146–4155, Sept 2013.

[14] M. S. Makowski, “A canonical switched capacitor dc-dc converter,” in 2014 IEEE 15th
Workshop on Control and Modeling for Power Electronics (COMPEL), June 2014, pp.
1–8.

[15] Y. Mahnashi and F. Peng, “Systematic approach to optimal sc converter synthesis for
multi voltage-gain-ratio applications,” in 2017 IEEE Workshop on Control and Modeling
for Power Electronics (COMPEL), 2017, pp. 1–5.

[16] D. Ma and R. Bondade, Reconfigurable Switched-Capacitor Power Converters. New
York, NY: Springer New York, 2013. [Online]. Available: http://dx.doi.org/10.1007/
978-1-4614-4187-8 5

[17] F. Su and W. H. Ki, “Design strategy for step-up charge pumps with variable inte-
ger conversion ratios,” IEEE Transactions on Circuits and Systems II: Express Briefs,
vol. 54, no. 5, pp. 417–421, May 2007.

[18] R. Karadi, “Synthesis of switched-capacitor power converters: An iterative algorithm,”
in 2015 IEEE 16th Workshop on Control and Modeling for Power Electronics (COM-
PEL), July 2015, pp. 1–4.

[19] C.-W. Kok, O.-Y. Wong, W.-S. Tam, and H. Wong, “Design strategy for two-phase
switched capacitor step-up charge pump,” in 2009 IEEE International Conference of
Electron Devices and Solid-State Circuits (EDSSC), Dec 2009, pp. 423–428.

103

[20] H. Taghizadeh, A. Cross, R. Whitehouse, and C. Barker, “An algorithm for automat-
ically calculating component current ratings in switched- capacitor dc-dc converters,”
IEEE Access, vol. PP, no. 99, pp. 1–1, 2018.

[21] D. Kilani, M. Alhawari, B. Mohammad, H. Saleh, and M. Ismail, “An efficient switched-
capacitor dc-dc buck converter for self-powered wearable electronics,” IEEE Transac-
tions on Circuits and Systems I: Regular Papers, vol. 63, no. 10, pp. 1557–1566, Oct
2016.

[22] B. Wu, L. Wang, L. Yang, K. M. Smedley, and S. Singer, “Comparative analysis of
steady-state models for a switched capacitor converter,” IEEE Transactions on Power
Electronics, vol. 32, no. 2, pp. 1186–1197, Feb 2017.

[23] M. D. Seeman, “A design methodology for switched-capacitor dc-dc converters,” Ph.D.

dissertation, University of California, Berkeley, 2009.

[24] M. D. Seeman and S. R. Sanders, “Analysis and optimization of switched-capacitor dc-
dc converters,” IEEE Transactions on Power Electronics, vol. 23, no. 2, pp. 841–851,
March 2008.

[25] B. Arntzen and D. Maksimovic, “Switched-capacitor dc/dc converters with resonant
gate drive,” IEEE Transactions on Power Electronics, vol. 13, no. 5, pp. 892–902, Sep
1998.

[26] G. Villar-Pique, H. J. Bergveld, E. Alarcon, G. V. Pique, H. J. Bergveld, and E. Alarcon,
“Survey and Benchmark of Fully Integrated Switching Power Converters : Switched-
Capacitor Versus Inductive Approach,” IEEE Transactions on Power Electronics,
vol. 28, no. 9, pp. 4156–4167, 2013.

[27] J. D. Vos, D. Flandre, and D. Bol, “A sizing methodology for on-chip switched-capacitor
dc/dc converters,” IEEE Transactions on Circuits and Systems I: Regular Papers,
vol. 61, no. 5, pp. 1597–1606, May 2014.

[28] Y. Lei and R. C. N. Pilawa-Podgurski, “Analysis of switched-capacitor dc-dc converters
in soft-charging operation,” in 2013 IEEE 14th Workshop on Control and Modeling for
Power Electronics (COMPEL), June 2013, pp. 1–7.

[29] H. Meyvaert, A. Sarafianos, N. Butzen, and M. Steyaert, “Monolithic switched-capacitor
dc-dc towards high voltage conversion ratios,” in 2014 IEEE 15th Workshop on Control
and Modeling for Power Electronics (COMPEL), June 2014, pp. 1–5.

104

[30] H. Meyvaert, M. Steyaert et al., High-ratio voltage conversion in CMOS for efficient

mains-connected standby. Springer, 2016.

[31] Online:, http://www.yarpiz.com. [Online]. Available: http://www.yarpiz.com

[32] M. M. Mano, Digital design. EBSCO Publishing, Inc., 2002.

[33] Online:, http://www.mosis.com/vendors/view/on-semiconductor/c5.

[Online]. Avail-

able: http://www.mosis.com/vendors/view/on-semiconductor/c5

[34] B. Razavi and , Design of analog CMOS integrated circuits. McGraw-Hill Education,

2000.

[35] J. Lu, A. Y. Kovalgin, K. H. M. van der Werf, R. E. I. Schropp, and J. Schmitz,
“Integration of solar cells on top of cmos chips part i: a-si solar cells,” IEEE Transactions
on Electron Devices, vol. 58, no. 7, pp. 2014–2021, July 2011.

[36] S. Ghosh, H. T. Wang, and W. D. Leon-Salas, “A circuit for energy harvesting using on-
chip solar cells,” IEEE Transactions on Power Electronics, vol. 29, no. 9, pp. 4658–4671,
Sept 2014.

[37] G. M. Pour, M. K. Benyhesan, and W. D. Leon-Salas, “Energy harvesting using sub-
strate photodiodes,” IEEE Transactions on Circuits and Systems II: Express Briefs,
vol. 61, no. 7, pp. 501–505, July 2014.

[38] C. Steffan, P. Greiner, B. Deutschmann, C. Kollegger, and G. Holweg, “Energy har-
vesting with on-chip solar cells and integrated dc/dc converter,” in 2015 45th European
Solid State Device Research Conference (ESSDERC), Sept 2015, pp. 142–145.

[39] S. Ayazian, V. A. Akhavan, E. Soenen, and A. Hassibi, “A photovoltaic-driven and
energy-autonomous cmos implantable sensor,” IEEE Transactions on Biomedical Cir-
cuits and Systems, vol. 6, no. 4, pp. 336–343, Aug 2012.

[40] Z. Chen, M. K. Law, P. I. Mak, and R. P. Martins, “A single-chip solar energy harvesting
ic using integrated photodiodes for biomedical implant applications,” IEEE Transac-
tions on Biomedical Circuits and Systems, vol. 11, no. 1, pp. 44–53, Feb 2017.

[41] N. J. Guilar, T. J. Kleeburg, A. Chen, D. R. Yankelevich, and R. Amirtharajah, “In-
tegrated solar energy harvesting and storage,” IEEE Transactions on Very Large Scale
Integration (VLSI) Systems, vol. 17, no. 5, pp. 627–637, May 2009.

105

[42] S. Ghosh, “Energy and data conversion circuits for low power sensory systems,” Ph.D.

dissertation, University of Missouri-Kansas City, 2013.

[43] J. D. Meindl and J. A. Davis, “The fundamental limit on binary switching energy for
terascale integration (tsi),” IEEE Journal of Solid-State Circuits, vol. 35, no. 10, pp.
1515–1516, Oct 2000.

[44] C. C. Enz, F. Krummenacher, and E. A. Vittoz, “An analytical mos transistor model
valid in all regions of operation and dedicated to low-voltage and low-current applica-
tions,” Analog integrated circuits and signal processing, vol. 8, no. 1, pp. 83–114, 1995.

[45] Y. K. Ramadass and A. P. Chandrakasan, “A battery-less thermoelectric energy harvest-
ing interface circuit with 35 mv startup voltage,” IEEE Journal of Solid-State Circuits,
vol. 46, no. 1, pp. 333–341, Jan 2011.

[46] Y. K. Tan and S. K. Panda, “Energy harvesting from hybrid indoor ambient light
and thermal energy sources for enhanced performance of wireless sensor nodes,” IEEE
Transactions on Industrial Electronics, vol. 58, no. 9, pp. 4424–4435, Sept 2011.

[47] R. D. Prabha, G. A. Rincn-Mora, and S. Kim, “Harvesting circuits for miniaturized
photovoltaic cells,” in 2011 IEEE International Symposium of Circuits and Systems
(ISCAS), May 2011, pp. 309–312.

[48] Z. Chen, M. K. Law, P. I. Mak, and R. P. Martins, “A single-chip solar energy harvesting
ic using integrated photodiodes for biomedical implant applications,” IEEE Transac-
tions on Biomedical Circuits and Systems, vol. 11, no. 1, pp. 44–53, Feb 2017.

[49] X. Liu and E. Snchez-Sinencio, “An 86% efficiency 12µ w self-sustaining pv energy har-
vesting system with hysteresis regulation and time-domain mppt for iot smart nodes,”
IEEE Journal of Solid-State Circuits, vol. 50, no. 6, pp. 1424–1437, June 2015.

[50] J. Kim, P. K. T. Mok, and C. Kim, “A 0.15 v input energy harvesting charge pump
with dynamic body biasing and adaptive dead-time for efficiency improvement,” IEEE
Journal of Solid-State Circuits, vol. 50, no. 2, pp. 414–425, Feb 2015.

[51] P.-H. Chen, K. Ishida, X. Zhang, Y. Okuma, Y. Ryu, M. Takamiya, and T. Sakurai,
“0.18-v input charge pump with forward body biasing in startup circuit using 65nm
cmos,” in IEEE Custom Integrated Circuits Conference 2010, Sept 2010, pp. 1–4.

[52] Y. K. Teh and P. K. T. Mok, “Design consideration of recent advanced low-voltage cmos
boost converter for energy harvesting,” in 2015 European Conference on Circuit Theory
and Design (ECCTD), Aug 2015, pp. 1–4.

106

[53] Z. Hameed and K. Moez, “Hybrid forward and backward threshold-compensated rf-
dc power converter for rf energy harvesting,” IEEE Journal on Emerging and Selected
Topics in Circuits and Systems, vol. 4, no. 3, pp. 335–343, Sept 2014.

[54] A. Cabrini, L. Gobbi, and G. Torelli, “Voltage gain analysis of integrated fibonacci-like
charge pumps for low power applications,” IEEE Transactions on Circuits and Systems
II: Express Briefs, vol. 54, no. 11, pp. 929–933, Nov 2007.

[55] O. Y. Wong, H. Wong, W. S. Tam, and C. W. Kok, “Parasitic capacitance effect on the
performance of two-phase switched-capacitor dc-dc converters,” IET Power Electronics,
vol. 8, no. 7, pp. 1195–1208, 2015.

[56] H. Meyvaert, T. V. Breussegem, and M. Steyaert, “A 1.65 w fully integrated 90 nm bulk
cmos capacitive dc-dc converter with intrinsic charge recycling,” IEEE Transactions on
Power Electronics, vol. 28, no. 9, pp. 4327–4334, Sept 2013.

[57] Y. K. Ramadass and A. P. Chandrakasan, “Voltage scalable switched capacitor dc-dc
converter for ultra-low-power on-chip applications,” in 2007 IEEE Power Electronics
Specialists Conference, June 2007, pp. 2353–2359.

[58] T. Tong, X. Zhang, W. Kim, D. Brooks, and G. Y. Wei, “A fully integrated battery-
connected switched-capacitor 4:1 voltage regulator with 70% peak efficiency using
bottom-plate charge recycling,” in Proceedings of the IEEE 2013 Custom Integrated
Circuits Conference, Sept 2013, pp. 1–4.

[59] J. H. Tsai, S. A. Ko, C. W. Wang, Y. C. Yen, H. H. Wang, P. C. Huang, P. H. Lan, and
M. H. Shen, “A 1 v input, 3 v-to-6 v output, 58%-efficient integrated charge pump with
a hybrid topology for area reduction and an improved efficiency by using parasitics,”
IEEE Journal of Solid-State Circuits, vol. 50, no. 11, pp. 2533–2548, Nov 2015.

[60] P. C. Lisboa, P. Prez-Nicoli, F. Veirano, and F. Silveira, “General top/bottom-plate
charge recycling technique for integrated switched capacitor dc-dc converters,” IEEE
Transactions on Circuits and Systems I: Regular Papers, vol. 63, no. 4, pp. 470–481,
April 2016.

[61] T. M. Andersen, F. Krismer, J. W. Kolar, T. Toifl, C. Menolfi, L. Kull, T. Morf,
M. Kossel, M. Brndli, and P. A. Francese, “Modeling and pareto optimization of on-
chip switched capacitor converters,” IEEE Transactions on Power Electronics, vol. 32,
no. 1, pp. 363–377, Jan 2017.

107

[62] T. Tanzawa, “On two-phase switched-capacitor multipliers with minimum circuit area,”
IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 57, no. 10, pp. 2602–
2608, Oct 2010.

[63] T. M. V. Breussegem, M. Wens, E. Geukens, D. Geys, and M. S. J. Steyaert, “Area-
driven optimisation of switched-capacitor dc/dc converters,” Electronics Letters, vol. 44,
no. 25, pp. 1488–1490, December 2008.

[64] R. J. Baker, CMOS: circuit design, layout, and simulation. John Wiley & Sons, 2008,

vol. 1.

[65] P. E. Allen and D. R. Holberg, CMOS analog circuit design. Oxford Univ. Press, 2002.

[66] J. He, S. Zhan, D. Chen, and R. L. Geiger, “Analyses of static and dynamic random
offset voltages in dynamic comparators,” IEEE Transactions on Circuits and Systems
I: Regular Papers, vol. 56, no. 5, pp. 911–919, May 2009.

[67] T. W. Matthews and P. L. Heedley, “A simulation method for accurately determining
dc and dynamic offsets in comparators,” in 48th Midwest Symposium on Circuits and
Systems, 2005., Aug 2005, pp. 1815–1818 Vol. 2.

[68] B. Gosselin and M. Sawan, “An ultra low-power cmos automatic action potential de-
tector,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 17,
no. 4, pp. 346–353, Aug 2009.

[69] L. Yan, P. Harpe, V. R. Pamula, M. Osawa, Y. Harada, K. Tamiya, C. V. Hoof, and
R. F. Yazicioglu, “A 680 na ecg acquisition ic for leadless pacemaker applications,”
IEEE Transactions on Biomedical Circuits and Systems, vol. 8, no. 6, pp. 779–786, Dec
2014.

[70] Y. C. Shih, T. Shen, and B. P. Otis, “A 2.3 mu w wireless intraocular pres-
sure/temperature monitor,” IEEE Journal of Solid-State Circuits, vol. 46, no. 11, pp.
2592–2601, Nov 2011.

[71] C. Hambeck, S. Mahlknecht, and T. Herndl, “A 2.4 µw wake-up receiver for wireless sen-
sor nodes with −71dbm sensitivity,” in 2011 IEEE International Symposium of Circuits
and Systems (ISCAS), May 2011, pp. 534–537.

[72] J. Choi, S. Park, J. Cho, and E. Yoon, “A 1.36 µw adaptive cmos image sensor with
reconfigurable modes of operation from available energy/illumination for distributed
wireless sensor network,” in 2012 IEEE International Solid-State Circuits Conference,
Feb 2012, pp. 112–114.

108

[73] Y. C. Shih and B. P. Otis, “An inductorless dc-dc converter for energy harvesting with
a 1.2µw bandgap-referenced output controller,” IEEE Transactions on Circuits and
Systems II: Express Briefs, vol. 58, no. 12, pp. 832–836, Dec 2011.

[74] D. Rozgi and D. Markovi, “A miniaturized 0.78-mw/cm2 autonomous thermoelectric
energy-harvesting platform for biomedical sensors,” IEEE Transactions on Biomedical
Circuits and Systems, vol. 11, no. 4, pp. 773–783, Aug 2017.

[75] I. Vaisband, M. Saadat, and B. Murmann, “A closed-loop reconfigurable switched-
capacitor dc-dc converter for sub-mw energy harvesting applications,” IEEE Transac-
tions on Circuits and Systems I: Regular Papers, vol. 62, no. 2, pp. 385–394, Feb 2015.

[76] S. Carreon-Bautista, L. Huang, and E. Sanchez-Sinencio, “An autonomous energy har-
vesting power management unit with digital regulation for iot applications,” IEEE Jour-
nal of Solid-State Circuits, vol. 51, no. 6, pp. 1457–1474, June 2016.

[77] Y. Yang, Y. Zhao, and G. Dong, “A multi-output on-chip switched-capacitor dcdc
converter with unequal flying capacitors for different power modes,” Journal of
Circuits, Systems and Computers, vol. 24, no. 04, p. 1550051, 2015. [Online]. Available:
http://www.worldscientific.com/doi/abs/10.1142/S0218126615500516

[78] N. Butzen and M. S. J. Steyaert, “Mimo switched-capacitor dc-dc converters using only
parasitic capacitances through scalable parasitic charge redistribution,” IEEE Journal
of Solid-State Circuits, vol. 52, no. 7, pp. 1814–1824, July 2017.

[79] T. M. Andersen, F. Krismer, J. W. Kolar, T. Toifl, C. Menolfi, L. Kull, T. Morf,
M. Kossel, M. Brndli, and P. A. Francese, “A 10 w on-chip switched capacitor volt-
age regulator with feedforward regulation capability for granular microprocessor power
delivery,” IEEE Transactions on Power Electronics, vol. 32, no. 1, pp. 378–393, Jan
2017.

[80] Y. Lu, J. Jiang, and W. H. Ki, “Design considerations of distributed and centralized
switched-capacitor converters for power supply on-chip,” IEEE Journal of Emerging
and Selected Topics in Power Electronics, vol. PP, no. 99, pp. 1–1, 2017.

[81] P. Zhou, D. Jiao, C. H. Kim, and S. S. Sapatnekar, “Exploration of on-chip switched-
capacitor dc-dc converter for multicore processors using a distributed power delivery
network,” in 2011 IEEE Custom Integrated Circuits Conference (CICC), Sept 2011, pp.
1–4.

[82] P. Huang, C. Lethien, S. Pinaud, K. Brousse, R. Laloo, V. Turq, M. Respaud,
and

A. Demorti`ere, B. Daffos, P. L. Taberna, B. Chaudret, Y. Gogotsi,

109

P. Simon, “On-chip and freestanding elastic carbon films for micro-supercapacitors,”
Science, vol. 351, no. 6274, pp. 691–695,
http:
//science.sciencemag.org/content/351/6274/691

[Online]. Available:

2016.

[83] K. Grigoras, J. Keskinen, L. Grnberg, E. Yli-Rantala, S. Laakso, H. Vlimki,
P. Kauranen, J. Ahopelto, and M. Prunnila, “Conformal titanium nitride in
a porous
in-chip supercapacitors,” Nano
Energy, vol. 26, no. Supplement C, pp. 340 – 345, 2016.
[Online]. Available:
http://www.sciencedirect.com/science/article/pii/S2211285516300842

silicon matrix: A nanomaterial

for

[84] H. Lin, W. C. Chan, W. K. Lee, Z. Chen, M. Chan, and M. Zhang, “High-current
drivability fibonacci charge pump with connect-point-shift enhancement,” IEEE Trans-
actions on Very Large Scale Integration (VLSI) Systems, vol. 25, no. 7, pp. 2164–2173,
July 2017.

[85] H. Lin and W. C. Chan and W. K. Lee and Z. Chen and M. Chan and M. Zhang, “A
high conversion ratio component-efficient charge pump for display drivers,” Journal of
Display Technology, vol. 12, no. 10, pp. 1057–1063, Oct 2016.

[86] Y. Ismail, H. Lee, S. Pamarti, and C. K. K. Yang, “A 36-v 49% efficient hybrid charge
pump in nanometer-scale bulk cmos technology,” IEEE Journal of Solid-State Circuits,
vol. 52, no. 3, pp. 781–798, March 2017.

[87] D. Lutz, P. Renz, and B. Wicht, “An integrated 3mw 120 / 230v ac mains micro power
supply,” IEEE Journal of Emerging and Selected Topics in Power Electronics, vol. PP,
no. 99, pp. 1–1, 2018.

110