Practical Implementation of Flying Capacitor BasedMultilevel Converters On-Chip and a Practical
Single-Mode Controller with Near Minimum DeviationTransient Response
by
Tom Moiannou
A thesis submitted in conformity with the requirementsfor the degree of Master of Applied Science
The Edward S. Rogers Sr. Department of Electrical and ComputerEngineering
University of Toronto
c© Copyright 2018 by Tom Moiannou
Abstract
Practical Implementation of Flying Capacitor Based Multilevel Converters On-Chip and
a Practical Single-Mode Controller with Near Minimum Deviation Transient Response
Tom Moiannou
Master of Applied Science
The Edward S. Rogers Sr. Department of Electrical and Computer Engineering
University of Toronto
2018
The work presented in this thesis focuses on the practical implementation of techniques
to reduce the volume of the output filter capacitor and inductor. The first area of focus
will be the on-chip implementation of flying-capacitor based multilevel converters, which
have been shown to reduce inductor volume. One such converter, the 7 Switch Flying
Capacitor (7SFC) Buck, is implemented with gate driving on-chip. In doing so, floorplan-
ning and layout strategies for multi-level converters were developed and are described.
As compared to conventional solutions, the volume is reduced while achieving better ef-
ficiency for most operating points. Also presented is a simplified single-mode controller
with near minimum deviation transient response for output capacitor volume reduction.
This controller achieves similar volume reduction as compared to multi-mode solutions
while eliminating dual-mode operation and associated toggling and issues. This con-
troller is demonstrated in a single-phase buck converter based prototype with extensions
to other topologies proposed.
ii
Acknowledgements
I would like to give thanks to my labmates through the past 2 and a half years. Parth,
Maryam, Behzad, Ahsan, Amr, Shadi, Abrar, Mia, Tim, Nenad, Samuel, Yuqing, Jas-
mine, Michael, Gianluca, Ivan, Sandy, and Basil. I have learned a great deal from all of
you and have grown as an engineer and as a person as a result. I’d also like to thank
Zhe for being a patient friend and a helpful colleague. I would also like to thank group
alumnus Aleksandar Radic for the guidance he has provided me with. I’d like to give
special thanks to Lindsey for being both an intellectual and emotional boost. I’d also
like to give special thanks to my colleague Andrija Stupar for his close cooperation and
support. Special thanks are also extended to Mahmoud Shousha and Martin Haug of
Wurth Elektronik for closely supporting my project and providing much needed guid-
ance at critical intervals. I would like to thank Wurth Elektronik generally for financially
backing this work. Lastly I would like to give extra special thanks to my supervisor
professor Aleksandar Prodic who has helped me tremendously during my time here. His
guidance and the laboratory environment he has worked hard to create has enabled me
to grow and become a substantially more effective engineer. I feel the past two years of
my life have been without question the most valuable ones thus far both professionally
and personally and I do not feel I could have grown in this way anywhere else.
iii
Contents
Acknowledgements iii
Table of Contents iv
List of Tables vii
List of Figures viii
List of Abbreviations xiii
1 Introduction 1
1.1 Fundamentals of Switch Mode Power Supplies . . . . . . . . . . . . . . . 2
1.2 Control of SMPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Prior Art and Motivation 7
2.1 Topologies for Volume Reduction . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Interleaved Buck Converter . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Three-Level Buck Converter . . . . . . . . . . . . . . . . . . . . . 11
2.1.3 High Step-Down Interleaved Buck Converter . . . . . . . . . . . . 14
2.1.4 7 Switch Flying Capacitor (7SFC) Buck . . . . . . . . . . . . . . 16
2.2 Review of Control Methods . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.1 Voltage Mode Control . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Fast Transient Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.1 Time-Optimal Controllers . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Minimum Deviation Controllers . . . . . . . . . . . . . . . . . . . 21
iv
3 Integrated Circuit Implementation of Emerging Topologies 23
3.1 Challenges Pertaining to Integration of Modern Topologies . . . . . . . . 25
3.1.1 Losses in Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.2 Layout Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Power stage Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Total Area Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.2 Realistic Technology Parameters for Area Selection . . . . . . . . 27
3.2.3 Transistor Sizing . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Floorplanning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Top Metal Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Gate Driving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5.1 Low Side Gate Driver . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5.2 High Side Gate Driver . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5.3 Gate Driver Layout . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6 Top-Level Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.7 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.7.1 Switching Waveforms . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.7.2 Conventional Solutions and Passive Volume Comparison . . . . . 45
3.7.3 Efficiency Results and comparison . . . . . . . . . . . . . . . . . . 46
4 Single-Mode controller with Near Minimum Deviation Transient Re-
sponse 50
4.1 Review of Fast Transient Controllers . . . . . . . . . . . . . . . . . . . . 51
4.2 Principle Of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2.1 Analysis of Switching Period Variation . . . . . . . . . . . . . . . 56
4.2.2 Stability for Small Perturbations . . . . . . . . . . . . . . . . . . 57
4.2.3 Extension to Multi-Phase Operation . . . . . . . . . . . . . . . . 58
4.2.4 Extension to Indirect Energy Transfer Converters . . . . . . . . . 59
4.3 Practical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4 Experimental and Simulation Results . . . . . . . . . . . . . . . . . . . . 62
5 Conclusions and Future Work 66
5.1 Integrated Implementation of Emerging Topologies . . . . . . . . . . . . 66
5.1.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.2 A Single-Mode Load-Tracking Voltage Mode Controller with Near Mini-
mum Deviation Transient Response . . . . . . . . . . . . . . . . . . . . . 67
5.2.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
v
A Bootstrapping Scheme 68
B Additional waveforms for IC 70
C Efficiency Curves 76
D Alternate Level Shifter Topologies 78
vi
List of Tables
3.1 Device Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Driver Chain Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Level Shifter Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4 Inductor Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5 Other Important Components . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6 Volume Comparison Between 7SFC and Similar Products . . . . . . . . . 45
3.7 PCB Area Comparison Between 7SFC and Similar Products . . . . . . . 46
4.1 Parameters of fast transient prototype . . . . . . . . . . . . . . . . . . . 62
vii
List of Figures
1.1 Taken from IPAD teardown. Shown highlighted in green are passives re-
lated to power management. Shown in blue are silicon devices related to
power management [1]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Schematic of a buck converter in a simple voltage mode control loop . . . 2
1.3 Simplified schematic of linear regulator (LDO), voltage divider formed
between transistor and load, transistor effective resistance adjusted by op-
amp such that load equals reference . . . . . . . . . . . . . . . . . . . . . 3
1.4 Two to one switched-capacitor circuit for step down . . . . . . . . . . . . 4
2.1 Schematic of an interleaved buck converter . . . . . . . . . . . . . . . . . 8
2.2 States for conversion ratios less than 0.5 . . . . . . . . . . . . . . . . . . 8
2.3 Switching waveforms for conversion ratios less than 0.5 . . . . . . . . . . 9
2.4 Interleaved buck converter switching states for conversion ratios greater
than 0.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Switching waveforms of an interleaved buck converter for conversion ratios
greater than 0.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6 Schematic of a 3-Level Buck . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.7 Switching states of a 3-level buck converter . . . . . . . . . . . . . . . . . 13
2.8 Switching waveforms of 3-level buck . . . . . . . . . . . . . . . . . . . . . 13
2.9 Schematic of the High step-down buck . . . . . . . . . . . . . . . . . . . 14
2.10 Switching states of HSD Buck . . . . . . . . . . . . . . . . . . . . . . . . 15
2.11 Switching waveforms of HSD buck converter . . . . . . . . . . . . . . . . 15
2.12 Schematic of the 7SFC. Adapted from schematic in [2] . . . . . . . . . . 17
2.13 Switching states and waveforms for 7sfc as 2-phase buck. Taken from [2] 17
2.14 Switching states and waveforms for the 7sfc operating as a three level buck.
Taken from [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.15 Switching states and waveforms for 7sfc as a 2-phase HSD buck. Taken
from [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
viii
2.16 Switching waveform of time-optimal transient response showing recovery
of lost charge with one switching action . . . . . . . . . . . . . . . . . . . 21
2.17 Waveforms for minimum deviation style controller correcting inductor cur-
rent within one cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1 A: Simplified top level schematic of IC developed, containing power-stage
and gate drivers. High side drivers shown for Q1,3,4,5,7, low-side drivers
from Q2,6 B: Schematic of 7SFC showing power transistors only, device
naming for power transistors corresponds to devices shown in A . . . . . 24
3.2 A: Layout of transistor ignoring maximum density rules for active layer.
Only valid for small devices. B) Layout of device considering maximum
density rules for active layer, device broken up into several smaller devices
and spaced apart such that the amount of the active layer in the given
area is smaller than the density limit. Total device width equal to device
shown in A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Pie chart showing area of transistors . . . . . . . . . . . . . . . . . . . . 29
3.4 Graph showing peak currents on interconnect between transistors and pads 31
3.5 A: Floorplan, shows the placement of the switches on the die B: Shows
how critical loops may be placed on a PCB . . . . . . . . . . . . . . . . . 32
3.6 Shows the metal thickness of the technology utilized. The top metal layer
is 8x thicker than the low metal layer, and can handle 8x more current. It
is rated for 8 mA/μm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.7 A: Shows the area where top metal usage is reserved for specific nets. For
instance, the area colored red is where top metal usage is assigned to the
net labelled IND2. The brown bars indicate where nets pads are connected
using lower metal layers exclusively. B: shows the corresponding color
coded nets in the schematic. . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.8 Block diagram showing schematic of low-side gate driver . . . . . . . . . 35
3.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.10 Showing function of high side driver, when on, output must remain 5 V
above source voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.11 Selected level shifter using cross-coupled topology . . . . . . . . . . . . . 39
3.12 Process and temperature simulations for the level-shifter operated for
Vs/VB of 0/5 and 48/53 . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.13 Layout of gate driver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.14 Top Level layout of 7SFC including gate drivers and transistors . . . . . 41
ix
3.15 Switches and gate drivers boxed in using black, pads shown in red . . . . 41
3.16 Photo of chip in package with lid removed . . . . . . . . . . . . . . . . . 42
3.17 Waveforms for 7SFC operating as a HSD Buck, showing inductor currents
and switching node voltages. Inductor currents measured with same offset
to show interleaved nature and equal current sharing. Ch1: Switching
node 1 voltage, 5 V/div. Ch2: Switching node 2 voltage, 5 V/div. Ch3:
Current of L1, 500 mA/div. Ch4: Current of L2, 500 mA/div. The time
scale is 500 ns/div. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.18 Waveforms for 7SFC operating as 3-Level Buck, showing inductor currents
and switching node voltage,as inductor waveforms are nearly identical dif-
ferent offset was used to show difference clearly. Ch1: Switching node 1
voltage, 10 V/div. Ch2: Switching node 2 voltage, 10 V/div. Ch3: Cur-
rent of L1, 500 mA/div. Ch4: Current of L2, 500 mA/div. The time scale
is 500 ns/div . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.19 Waveforms for 7SFC operating as 2-phase interleaved buck, showing induc-
tor currents and switching node voltages. Ch1: Switching node 1 voltage,
10 V/div. Ch2: Switching node 2 voltage, 10 V/div. Ch3: Current of L1,
500 mA/div. Ch4: Current of L2, 500 mA/div. The time scale is 500 ns/div 45
3.20 Efficiency curves for the 7SFC using 2.2μH inductors in a 1210 package.
Title indicates input and output voltages as well as operating mode (3LB
= 3 Level Buck, HSD = High Step Down Buck). Frequencies indicated in
legend. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.21 Efficiency curves for the 7SFC using the 2.2μH inductors in an 0806 LR
package. Title indicates input and output voltages as well as operating
mode (3LB = 3 Level Buck, HSD = High Step Down Buck). Frequencies
indicated in legend. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1 Block diagram of the proposed controller used in a single-phase buck con-
verter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2 Ideal waveforms of the proposed control scheme operating in steady-state.
From top to bottom are the output voltage and nominal output voltage
on the same axis, the inductor current and load current on the same axis,
the capacitor current, the output of the polarity detector, gating signal of
Q1, and gating signal of Q2. . . . . . . . . . . . . . . . . . . . . . . . . . 53
x
4.3 Ideal waveforms of the proposed control scheme operating during a light-
to-heavy load step. From top to bottom are the output voltage and nomi-
nal output voltage on the same axis, the inductor current and load current
on the same axis, the capacitor current, the output of the polarity detector,
gating signal of Q1, and gating signal of Q2 . . . . . . . . . . . . . . . . . 55
4.4 Voltage recovery after load transient for single-mode controller . . . . . . 56
4.5 Showing response of controller in multi-phase case . . . . . . . . . . . . . 58
4.6 Controller modified to be used in a boost converter, note modulator re-
mains same, only difference is capacitor time constant estimator replaced
with 2 sensors and digital potentiometer . . . . . . . . . . . . . . . . . . 59
4.7 Showing waveforms for control scheme applied in boost converter . . . . 59
4.8 Structure of the RC time constant estimator and zero-crossing detection
circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.9 Block diagram of half-duty modulator . . . . . . . . . . . . . . . . . . . . 61
4.10 Simulation of controller for 4 A load step, depicting output voltage, in-
ductor current, and load current. . . . . . . . . . . . . . . . . . . . . . . 63
4.11 A: Oscilloscope screenshot showing output capacitor voltage and inductor
current after 4 A light to heavy load step, (ILoad 0.6 to 4.6A). Ch1 vout(t)
, 100 mV/div. Ch2 iL(t), 2 A/div. Timescale is 2 us/div. B: Zoomed out
version of A showing voltage recovery phase, timescale set to 500 us/div.
C: Identical loadstep, except going from heavy to light, D: zoomed out
loadstep shown in C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.12 A: Controller responding to consecutive load steps. Ch1 vout(t), 100 mV/div.
Ch2 iL(t), 1 A/div, timescale is 5 μs B: Reference step indicating con-
troller’s stability. Ch1 vout(t), 1 V/div, timescale is 1 ms . . . . . . . . . 64
4.13 A: Depicts a load-step for the control scheme applied to a 2-phase con-
verter. B: Depicts a load-step for the control scheme applied to a Boost
converter, showing the output voltage and scaled inductor current such as
would be seen at the output of the digital potentiometer. . . . . . . . . 65
A.1 The bootstrapping scheme used for power the gate drive circuits in the
designed power stage of the 7SFC . . . . . . . . . . . . . . . . . . . . . . 69
B.1 Showing gate waveforms for Q1 and Q2 as well as the two main switching
nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
B.2 Switching waveforms for HSD mode of operation, showing gate of switch
4 and 6, as well as switching node 2 . . . . . . . . . . . . . . . . . . . . . 71
xi
B.3 Switching waveforms for HSD, showing gate of switch 3, 1, and vsw1(t) . . 72
B.4 Switching waveforms for HSD, showing gates of switches 7 and 1 as well
as vsw1(t) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
B.5 Switching waveforms for HSD, showing gate of switch 5, 4, and vsw2(t) . . 74
B.6 Switching nodes and low-side gating signals when operating in HSD mode
for 42 V input voltage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
C.1 Additional efficiency curves using the inductors in the 0806 LR package . 76
C.2 Additional efficiency curves using using the inductors in the 1210 size package 77
D.1 Pulsed SR Latch based level-shifting topology . . . . . . . . . . . . . . . 79
D.2 Simulation showing output swing of pulsed SR latch across process and
temperature corners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
D.3 Modified level shifter to limit swing to less than 5 V . . . . . . . . . . . . 80
D.4 Simulation showing level shifter perforamance across several corners . . . 81
D.5 Schematic of bootstrapped level shifter . . . . . . . . . . . . . . . . . . . 82
xii
List of AbbreviationsSMPS Switch Mode Power Supply
PID Proportional Integral Derivative
CCP Current Correction Phase
RRP Ripple Reconstruction Phase
IC Integrated Circuit
7SFC 7 Switch Flying Capacitor
HSD High Step Down
3LB 3 Level Buck
IB Interleaved Buck
CCM Continuous Conduction Mode
DCM Discontinuous Conduction Mode
PFM Pulse Frequency Modulation
ESD Electrostatic Discharge
LDO Low Dropout Regulator
PoL Point of Load
CPM Current Programmed Mode Control
PCB Printed Circuit Board
xiii
Chapter 1
Introduction
Switch-Mode power supplies are virtually ubiquitous in modern consumer electronics.
DC-DC power supplies are used to regulate the electric power supplied to one or more
functional blocks. These supplies consume large amounts of PCB area and contribute
substantially to the total volume of many modern electronic devices. This can be seen
in Fig 1.1, using the results of an Ipad teardown taken from [1].
Figure 1.1: Taken from IPAD teardown. Shown highlighted in green are passives relatedto power management. Shown in blue are silicon devices related to power management[1].
Semiconductor portions of the power supplies are shaded in blue. Passives related to
power management are shaded in green. As can be seen, these components, particularly
the passives, consume large amounts of area. Emerging technologies present opportunities
to reduce the volume of such passive components. This includes new topologies to reduce
inductor volume and control schemes to reduce output filter capacitor volume. Doing so
creates room to add or improve other features on such devices, or otherwise permits the
shrinking of electronic devices altogether.
1
Chapter 1. Introduction 2
1.1 Fundamentals of Switch Mode Power Supplies
Switch-mode power supplies are popular choices for converting power from one form to
another. They consist of three basic components. A switching network, an output filter,
and a controller. Fig 1.2 depicts a buck converter in a voltage mode control loop as an
example. The switching network modulates the input of the filter, passing it rectangular
pulses equal to Vg and ground, of durations DTs and (1-D)Ts respectively. Where Ts
is the switching period and D the duty cycle. The output filter functions as a low pass
filter, causing the average value of the voltage to appear at the output. This value is
equal to VgDTs.
Figure 1.2: Schematic of a buck converter in a simple voltage mode control loop
The output filter components shown in Fig. 1.2 correspond to the passives shown
in Fig. 1.1, which were highlighted in green. The semiconductor portions of the buck
(i.e. the two transistors) are included the semiconductor portions of Fig. 1.1. Most
of the semiconductor portions of Fig. 1.1 are in the form of power management inte-
grated circuits, or PMICs. These integrate both transistors, as well as ancillary circuitry
such as gate drivers and control circuits into a single package. This reduces the volume
overhead associated with separately packaged transistors and is frequently done to im-
prove performance and avoid wasting space, and is very desirable in volume sensitive
applications.
Switch-mode power supplies compete with two existing technologies in the area of DC-
DC conversion. One example is the linear regulator which is shown in Fig. 1.3. Linear
regulators place a variable resistor (often realized as a MOSFET) in series with the load.
Chapter 1. Introduction 3
The resistance is varied such that the divider formed between it and the load provide
the load with the desired voltage. Unlike switch-mode power supplies this approach may
only be used to step-down a voltage, not step up one. Furthermore it is intrinsically
lossy, as all of the excess energy is burnt off resistively as opposed to being stored. Its
theoretical maximum efficiency is equal to the load voltage divided by the input voltage
[3]. Hence such regulators have been largely replaced by switch-mode power supplies in
most applications as these can theoretically achieve 100% efficiency for any conversion
ratio. Although, LDOs are still popular for low-step down (conversion ratios close to 1)
and low-power applications. They are also popular when extremely tight regulation of
the output voltage is required and/or very little supply noise can be tolerated. In such
cases they are often used downstream of an SMPS.
Figure 1.3: Simplified schematic of linear regulator (LDO), voltage divider formed be-tween transistor and load, transistor effective resistance adjusted by op-amp such thatload equals reference
SMPS also compete with switched-capacitor circuits which can multiply or divide a
voltage by a fixed amount. The schematic of a two to one step down switched capaictor
circuit is depicted in Fig. 1.4. Like SMPS, these can theoretically acheive very high
efficiencies [4]. Unlike SMPS however, they have difficulty providing tight regulation.
These circuits work by multiplying or dividing the input by a fixed number. Control
techniques can be applied to adjust the output, but this degrades efficiency considerably
[4]. Hence these converters are best suited for applications where the input is fixed or
very tight regulation of the output is not required.
Chapter 1. Introduction 4
Figure 1.4: Two to one switched-capacitor circuit for step down
The size of the components in power supplies is determined by a number of factors.
The size of the inductor is determined by both efficiency and EMI specifications, as
larger inductance values lead to reduced ripple and thus lower rms current (losses) and
less electromagnetic interference. The size of the output capacitors may be determined
by either the steady-state ripple or the transient response of the converter. In cases where
load transients are a concern, this is usually the dominant factor.
Emerging technologies present an opportunity to reduce the volume of switch-mode
power supplies. New topologies for power conversion permit reduced passive volume,
particularly that of the output filter inductors. Advanced control schemes offer the
ability to reduce the size of the output capacitors by improving the transient response of
converters, such that regulation requirements may be met using smaller capacitors.
1.2 Control of SMPS
Traditionally, switch-mode power supplies are controlled using a linear control scheme,
such as voltage-mode control. A block diagram for this control scheme was shown in
figure 1.2. The converter is placed in a loop. A reference voltage is provided to the error
generator, which compares it to the output voltage and generates an error signal. The
error is supplied to the linear compensator, typically a PI or PID. The compensator then
adjusts the duty cycle command, which is used to generate the appropriate pulse-width
Chapter 1. Introduction 5
modulated gating signals to control the output voltage.
This controller may be implemented in either an analog or mixed-signal manner. It
is increasingly popular to use digital compensators as they require less silicon area to im-
plement in modern technology, while also permitting more aggressive control techniques
(such as tuning coefficients based on operating conditions).
Control schemes such as the one described have less than ideal transient performance.
New control schemes which do not rely solely on linear compensators have been proposed
to achieve a more aggressive transient response. Such technologies include minimum de-
viation and time-optimal controllers. Minimum-deviation and time-optimal controllers
generally use dual-mode operation, having one mode for steady-state and one for load-
transients. This can lead to issues where repeated toggling occurs between modes de-
grading performance. These controllers will be discussed in greater detail in chapter
2.
1.3 Thesis Objectives
This thesis will discuss the practical implementation of emerging volume-reduction tech-
niques for switch-mode power supplies. Shown first will be the on-chip implementation
of a recently developed topology, the Seven Switch Flying Capacitor Buck, which is ca-
pable of reducing volume for comparable efficiency. This converter belongs to a class
of converters known as multi-level flying capacitor converters. As will be discussed in
chapter two, these converters show great promise for reducing inductor volume, but have
rarely been integrated on-chip. Integration of such converters is necessary for their wider
adoption in volume sensitive applications. There are many challenges to integrating such
converters which must be addressed. One of the main aims of this thesis is to address
these challenges. Some key objectives of this aspect of the thesis are listed below.
1. Discuss the sizing of devices in emerging topologies
2. Formulate and provide practical guidelines for the floorplanning and metal routing
of modern step-down topologies of high complexity
3. Show the suitability of low-power gate drive circuits
4. Demonstrate reduced volume for comparable efficiency when compared to IC mod-
ules employing the conventional buck converter
Chapter 1. Introduction 6
This thesis will also aim to address one of the main limitations associated with min-
imum deviation style controllers, that is, dual-mode operation. Proposed in this thesis
is a single-mode controller that offers near minimum deviation transient response. This
eliminates the risk of toggling issues and the instability that can result. This aspect of
the thesis will aim to do the following.
1. Describe algorithm for single-mode controller with near minimum deviation tran-
sient response
2. Demonstrate implementation of controller in a single-phase buck converter based
prototype
3. Propose extension to additional converter topologies
1.4 Thesis Overview
Chapter 2 will review prior art and technologies. First topological innovations will be
discussed. This will focus on “multilevel” converters, which enable a reduction in inductor
volume by reducing the voltage swing at the switching node of a SMPS. Afterwards, fast
transient control schemes which permit a reduction in the output capacitor volume will
be discussed.
Chapter 3 will look at the on-chip implementation of a multilevel converter. The
process will be explored in detail and efficiency results including comparison to existing
conventional products will be done. The design and layout of the power stage and gate
drivers will be discussed. Efficiency results measured from the fabricated chip will be
presented and discussed.
Chapter 4 will look at a new, practical, single-mode near minimum deviation control
scheme. Its implementation will be discussed, and experimental results for a buck con-
verter based prototype will be presented. The extension of this control scheme to other
topologies will also be discussed and simulation results will be presented.
Chapter 5 will present conclusions and discuss future work.
Chapter 2
Prior Art and Motivation
Many technologies have been developed for the purpose of reducing volume in switch-
mode power supplies. In this chapter, topological and control techniques will be reviewed.
2.1 Topologies for Volume Reduction
2.1.1 Interleaved Buck Converter
Numerous topologies have been devised to reduce the size of passive elements in switching
converters. One such group of topologies are phase-interleaved converters [5]. This
technique works by essentially splitting the switching network and inductor into several
parallel stages. The stages are operated out of phase such that they are charging and
discharging at different times. This reduces the amplitude of the output capacitor current,
thus reducing the converter’s ripple. This can reduce the output capacitor volume [6], [7].
This technique can also reduce the volume of the input filter by reducing the amplitude
of the current pulses drawn from this branch [6], [7]. Furthermore, the net reduction
in current ripple at both the input and the output results in lower RMS currents and
can lead to an improvement in efficiency [6], [7]. Lastly, phase interleaving can better
distribute losses in a converter and reduce local heating.
A 2-phase interleaved buck is shown in Fig 2.1. This converter can follow two different
sets of switching states. One set of switching states allows for conversion ratios less than
0.5, the other is for conversion ratios above it. For conversion ratios less than 0.5,
switching states are shown in 2.2 and switching waveforms are shown in Fig 2.3. For
greater than 0.5, the states and waveforms are shown in Figs 2.4 and 2.5 respectively.
7
Chapter 2. Prior Art and Motivation 8
Figure 2.1: Schematic of an interleaved buck converter
Figure 2.2: States for conversion ratios less than 0.5
Chapter 2. Prior Art and Motivation 9
Figure 2.3: Switching waveforms for conversion ratios less than 0.5
Chapter 2. Prior Art and Motivation 10
Figure 2.4: Interleaved buck converter switching states for conversion ratios greater than0.5
Figure 2.5: Switching waveforms of an interleaved buck converter for conversion ratiosgreater than 0.5
A single switching cycle consists of 4 states. For conversion ratios less than 0.5, when
in its first state, switches 1 and 4 are on, the others are off. Inductor L1 is charging up,
Chapter 2. Prior Art and Motivation 11
L2 is discharging. Upon entering its second state, switches 2 and 4 are on, both inductors
are discharging. Upon entering its third state, switches 2 and 3 are on, and inductor L2
is charging up. Its fourth state is a repeat of state 2.For conversion ratios greater than
0.5, states 1 and 3 have both inductors charging, states 2 and 4 set inductors L1 and L2
into rectification respectively.
2.1.2 Three-Level Buck Converter
Another technique which has emerged in the past 2 decades and is beginning to see
broader adoption is the usage of flying capacitor based multilevel converters. These
converters use one or more flying capacitors to reduce the voltage swing at the various
switching nodes. Hence the inductor volume can be reduced considerably while main-
taining the same inductor ripple frequency and peak current [8].
The schematic of a three-level buck is shown in Fig 2.6 [9]. It functions similarly to
a conventional buck converter. However, there are two additional switches and a flying
capacitor. The flying capacitor voltage is typically about one-half the input voltage,
although it may deviate from this value in the absence of a controller to keep it there.
Depending on start-up conditions, this converter may be implemented with switches rated
for one half of the input voltage, which consume substantially less area than switches
rated for the full input voltage. This can be better explained by looking at the Figure
of Merit (FOM) of the transistor. The most commonly used FOM is the product of its
on-resistance and gate charge, RDSON · QG. Typically, the FOM of transistor is related
to VDS in the following manner, FOM = αV 2DS, where α is a constant of proportionality
[8]. Hence, halving the voltage rating improves the FOM by a factor of 4. As a result,
it is theoretically possible to implement multilevel converters with no increase in silicon
area [8].
Chapter 2. Prior Art and Motivation 12
Figure 2.6: Schematic of a 3-Level Buck
The switching states of this converter are shown in Fig 2.7. Corresponding switching
waveforms for this converter are shown in Fig 2.8. During the first portion of the cycle
switches 1 and 3 are on, switches 2 and 4 are off. The input voltage is passed through the
flying capacitor to the output filter. As the flying capacitor voltage is equal to half the
input voltage the switching node sees the input voltage minus half of its value. During
phase 2, switch 4 is on, and all other switches are off, the switching need is zero. During
phase three, switches 2 and 4 are on. The flying capacitor voltage is passed to the output
filter. Phase four is identical to phase 2.
Chapter 2. Prior Art and Motivation 13
Figure 2.7: Switching states of a 3-level buck converter
Figure 2.8: Switching waveforms of 3-level buck
The conversion ratio of this converter is equal to DTs, where Ts is the period of time
it takes for all four phases to complete. DTs is equal to the duration of phases 1 and
3 individually (i.e. the sum of the 2 is 2DTs). As compared to a conventional buck
Chapter 2. Prior Art and Motivation 14
converter operating with an equal Ts, the inductor current ripple is half the value. This
allows one to use smaller inductors with all other things being kept constant.
2.1.3 High Step-Down Interleaved Buck Converter
Another multilevel flying capacitor based topology is the High Step-Down Interleaved
Buck Converter [10]. A schematic of this converter is shown in figure 2.9. Like the
3-level buck converter, this topology uses a flying capacitor whose voltage is equal to
that of one half the input voltage. This topology has several advantages over the 3-
level buck. First and foremost, as a multiphase converter, it will see reduced output
capacitor voltage ripple. As the inductance requirements are otherwise equal this does
not impact overall inductor volume, as its output filter is two inductors each rated for
half the current. Unlike the 3-level buck converter, the flying capacitor voltage does not
deviate significantly from half of the input voltage. This is because Inductor Volt-Second
Balance (IVSB) may only be satisfied in each inductor if the voltage is exactly half of the
input voltage (in the absence of losses). If the losses along each path of this converter are
not equivalent, a small voltage imbalance in the flying capacitor will appear to correct
them. However this is typically just a few percent of its nominal value. The switching
states of this converter are provided in Fig. 2.10 and switching waveforms in Fig. 2.11.
Figure 2.9: Schematic of the High step-down buck
Chapter 2. Prior Art and Motivation 15
Figure 2.10: Switching states of HSD Buck
Figure 2.11: Switching waveforms of HSD buck converter
This converter has a conversion ratio of D/2. Where D is defined as the of the
duration of phase 1 (which is equal to that of phase 3) divided by the switching period
Chapter 2. Prior Art and Motivation 16
Ts. This ties into yet another benefit of the double step-down buck converter. It has duty
cycle extension, which makes it useful for high-step down. As its duty cycle is longer,
it relaxes switching speed requirements for very high conversion ratios M(D) < 0.1. If
digital control is used, this also reduces requirements on the resolution of the DPWM.
This converter also has benefits over the conventional interleaved buck converter. As
each phase must have equal current in order for the flying capacitor voltage to settle to
half the input voltage, the inductor currents in each phase balance inherently. Whereas
there is nothing in a conventional interleaved converter that forces the currents in each
phase to match each other.
This converter does however have drawbacks when compared to the 3-level buck.
The inductor current ripple will be twice that of the 3-level buck for the same conversion
ratio. It also requires switch two be rated for the full input voltage regardless of the
startup condition. As its conversion ratio is D/2, and its maximum duty cycle is 0.5, it
is incapable of achieving conversion ratios greater than 0.25. which makes it unsuitable
for applications requiring wide output range or low conversion ratios (M(D) > 0.25).
2.1.4 7 Switch Flying Capacitor (7SFC) Buck
The 7SFC is a flying capacitor based topology that can be configured to operate as any
of the previously presented topologies [11]. A schematic of it is shown in Fig. 2.12. All
switches in it may be rated for half the input voltage. It selects its operating mode based
on the conversion ratio required. It typically operates as a 3-level buck, a high step-down
buck or a interleaved buck converter. This allows it to harness the powerful benefits of
the high step down topology while still being able to achieve wide-range operation.
When operating as a three-level buck, both inductors are ran in parallel and treated
like a single inductor. When operating as a double step down or interleaved buck it
operates just as those converters do as described above. One caveat is that if it is imple-
mented using half-rated switches it may only use the interleaved buck mode if operated
for an input voltage less than one half of its fully rated input voltage. When operating
as the three aforementioned converters the switching nodes and inductor currents will
behave as described for those converters. Switching states and waveforms for the three
different modes are provided in Figs. 2.13, 2.14, and 2.15.
Chapter 2. Prior Art and Motivation 17
Figure 2.12: Schematic of the 7SFC. Adapted from schematic in [2]
Figure 2.13: Switching states and waveforms for 7sfc as 2-phase buck. Taken from [2]
Chapter 2. Prior Art and Motivation 18
Figure 2.14: Switching states and waveforms for the 7sfc operating as a three level buck.Taken from [2]
Chapter 2. Prior Art and Motivation 19
Figure 2.15: Switching states and waveforms for 7sfc as a 2-phase HSD buck. Taken from[2]
Chapter 2. Prior Art and Motivation 20
2.2 Review of Control Methods
Numerous control methods have been developed to regulate the output voltage of SMPS
controllers and improve their load-transient response. This includes, steady-state con-
trollers, like voltage mode controllers, and fast transient controllers, like time-optimal
and minimum deviation style controllers. These will be discussed in this section.
2.2.1 Voltage Mode Control
A block diagram for a voltage mode controller was shown in Fig. 1.2. In essence a voltage
mode controller reads the output voltage and adjusts the duty cycle until it reaches the
desired value. It is the simplest control scheme typically used for power converters and is
highly flexible and easy to implement. However, voltage mode controllers cannot provide
the best transient response.
2.3 Fast Transient Controllers
In recent years, numerous fast load-transient response controllers have been proposed.
Examples include time-optimal [12],[13],[14],[15],[16] and minimum deviation controllers
[17],[18],[19]. For direct energy transfer converters, such as a buck converter, the usage of
such controllers results in what is practically the minimum achievable voltage deviation
for a given converter and load-step [11]. This presents the opportunity to drastically
reduce the output capacitor volume. Still, the presented approaches suffer from a few
drawbacks that have slowed down their wider adoption in particular targeted applications.
2.3.1 Time-Optimal Controllers
Time-optimal controllers are fast transient controllers that aim to recover lost output
voltage in the minimum amount of time possible. Fig. 2.16 shows the waveforms during a
transient. Normally, the converter operates using a linear control scheme, such as voltage
mode control. When a load transient is detected, the inductor current begins charging
up until the output voltage ceases to drop. Once this happens, a peak inductor current
value is calculated such that the lost voltage is recovered. This is done by determining
a peak value such that charge Q2 is equal to charge Q1. This means that the charge
deposited back into the capacitor by the increased inductor current is equal to that lost
during the transient. This returns the output voltage to its nominal value. The controller
then resumes normal operation.
Chapter 2. Prior Art and Motivation 21
Figure 2.16: Switching waveform of time-optimal transient response showing recovery oflost charge with one switching action
This control scheme provides the fastest possible transient response but suffers from
several problems. First off, it is computationally intensive, calculating the peak current
requires large amounts of digital circuitry. It also requires power hungry high speed
ADCs which consume substantial amounts of power and silicon area. Thirdly, it requires
the inductor used be rated according to the peak current the controller requires. Next,
it has difficulty responding to successive transients, which can lead to toggling between
its transient response and steady state control modes [20]. Losses and non-ideal effects
in real converters can also result in errors in its transient response and lead to toggling
problems [20]. Lastly it requires knowledge of converter to parameters to work properly
limiting its utility and making it susceptible to variations in these parameters [20].
2.3.2 Minimum Deviation Controllers
Minimum deviation controllers address many of the issues with time-optimal controllers
at the cost of a slower voltage recovery time. These controllers focus on correcting the
inductor current as quickly as possible during a load transient such that it matches
the load. Ideal waveforms for this control scheme are shown in 2.17. In the event of
a load transient, a minimum deviation controller simply forces the inductor current to
Chapter 2. Prior Art and Motivation 22
match that of the load. This avoids some of the issues of time-optimal converters. As
no calculations are required, the digital hardware requirements are eliminated. As the
inductor current is not ramped past the desired steady-state value, the inductor need only
be sized as per its steady state requirements. There is also no need to sense the inductor
current digitally, eliminating a high-cost ADC from the equation. This controller gives
the minimum achievable voltage deviation possible for a given load transient.
Figure 2.17: Waveforms for minimum deviation style controller correcting inductor cur-rent within one cycle
Chapter 3
Integrated Circuit Implementation
of Emerging Topologies
Emerging topologies have rarely been implemented on-chip. This thesis will show the
implementation of the 7SFC into a on-chip power module with integrated gate drivers.
In addition to conventional topics like gate driving, challenges that are unique to more
complex topologies will be discussed. One example is device sizing. As the converter
operates in many different modes, the manner in which silicon area need be assigned
is a more complicated topic than it is in other converters. Another challenge is floor-
planning. As many conduction paths exist, and different modes will see different paths
conducting large amounts of current, special care must be taken to optimize the arrange-
ment of devices. This is necessary to minimize the contribution of metallization to the
on-resistance of the devices, which as will be shown is a significant contributor, and to
meet electromigration requirements.
The power stage is designed as a wide-input, step-down, point-of-load module, which
must be able to interface with input BUS voltages ranging from 5V - 42V, outputting
voltages from 0.8 V to 12 V (but not exceeding the input voltage of the converter). The
maximum output current of this design is to be 2 A. A block diagram of the IC developed
is provided in Fig. 3.1, the functional blocks of this chip will be discussed in the sections
below. This operating range is desirable for consumers seeking a single product to deliver
power to a variety of loads from different bus voltages. Products with similar operating
ranges are compared to the implemented converter.
23
Chapter 3. IC Implementation of Emerging Topologies 24
Figure 3.1: A: Simplified top level schematic of IC developed, containing power-stageand gate drivers. High side drivers shown for Q1,3,4,5,7, low-side drivers from Q2,6 B:Schematic of 7SFC showing power transistors only, device naming for power transistorscorresponds to devices shown in A
Chapter 3. IC Implementation of Emerging Topologies 25
3.1 Challenges Pertaining to Integration of Modern
Topologies
There are many challenges that are intrinsic to the on-chip implementation of SMPS that
are exacerbated by the complexity of modern emerging topologies. One major goal of
this thesis is to address how to approach these challenges.
3.1.1 Losses in Devices
In a conventional buck converter, there are two transistors, a high-side, and a low-side
device. For conversion ratios closer to 1, the high-side device will conduct the majority
of the current. For conversion ratios closer to zero, the low-side device will conduct the
majority of the current. This is shown using the following expressions [21].
PHS =
√D · (I2OUT +
I2Ripple
12) ·RDSON (3.1)
PLS =
√(1 −D) · (I2OUT +
I2Ripple
12) ·RDSON (3.2)
Where PHS and PLS are the conduction losses of the high and low side devices respec-
tively. Hence the losses in each switch is strictly related to the duty cycle and thus the
conversion ratio. This is not true for the 7SFC. For instance Q7 will nominally conduct no
current in the HSD mode of operation for all conversion ratios. However, when operating
as a 3-Level Buck or as an interleaved buck, will experience current stresses similar to a
high-side switch in a conventional buck converter. Conduction losses in Q2 and Q3 are
directly proportional to 1-D in the interleaved buck mode of operation, but proportional
to Ts/2 + (1-D) in the 3-level buck and HSD modes of operation. This complicates the
sizing of these devices. The conduction losses of Q3 are shown in equations 3.3 and 3.4
for the interleaved buck and high-step down modes respectively.
PCOND Q3 IB =
√(1 −D) · ((0.5 · IOUT )2 +
I2Ripple
12) ·RDSON
(3.3)
Chapter 3. IC Implementation of Emerging Topologies 26
PCOND Q3 HSD =
√(D) · ((0.5 · IOUT )2 +
I2Ripple
12) ·RDSON
+2 ·
√(1 −D) · ((0.5 · IOUT )2 +
I2Ripple
12) ·RDSON
(3.4)
This problem is further complicated by gate-drive losses and Coss losses. As some
devices switch in some modes and not others, and Q6 will switch at twice the nominal
switching frequency in the 3-level buck mod of operation.
3.1.2 Layout Challenges
Related to the number of devices and different current stresses is the interconnect between
devices and pads. Unlike an on-chip buck converter, which contains two transistors,
interfacing to 3 power related external connections (input voltage, output filter, and
ground),the 7SFC, contains 7 devices, which interface with 7 external connections (two
separate ground ports, the input voltage, one port per inductor, and one port connecting
to either side of the flying-capacitor. Making matters worse there is the electromigration
phenomenon. Electromigration is a long-term failure mechanism which can lead to open
circuits forming on metal pathways due to the tendency of metal to “migrate” in the
direction of current flow [22]. There are maximum current density specifications which
need to be satisfied to ensure this does not happen within the devices expected lifetime
[22]. This is particularly challenging in the 7SFC due to the many current paths leading
to and from most devices, many of which experience high current in one mode but not
others. One example of this is the connection between Q1 and Q4. In the HSD and 3-
Level Buck modes of operation this path conducts no current. However in the interleaved
buck mode of operation this pathway can conduct up to the full-load current.
Another important thing to consider is the layout of the printed circuit board which
the IC will be soldered to. One important thing about this is minimizing the length and
loop area of the switching node loops. This is important to minimize parasitic inductance
and EMI related problems [23]. If possible the IC should be laid out in such a way that
the power path all be done on one layer so that inductance not be added by using vias
to jump layers [23].
Chapter 3. IC Implementation of Emerging Topologies 27
3.2 Power stage Design
3.2.1 Total Area Usage
The area required for the converter is based on that used by similar products. This figure
was provided to us by Wurth Elektronik who was sponsoring the project, and was set to
5mm2.
3.2.2 Realistic Technology Parameters for Area Selection
There are several parameters which are useful when sizing power transistors. These
parameters are listed as follows.
1. Rsp - The specific on-resistance of a given technology, that is the on-resistance of a
1mm2 transistor made in this techonology, given in mΩ·mm2
2. Coss density - the drain to source capacitance density of the technology used (pF/mm2)
3. Cin density - the gate to source capacitance density of the technology (pF/mm2)
The foundry provides nominal values for these parameters, however this number is not
very accurate as it does not take into account layout considerations. The figure provided
does not take into account layer density rules which limit the amount of active transistor
area that can be placed in a given area. Hence it is only accurate for very small devices.
This can be understood by looking at Fig 3.2
Chapter 3. IC Implementation of Emerging Topologies 28
Figure 3.2: A: Layout of transistor ignoring maximum density rules for active layer.Only valid for small devices. B) Layout of device considering maximum density rules foractive layer, device broken up into several smaller devices and spaced apart such thatthe amount of the active layer in the given area is smaller than the density limit. Totaldevice width equal to device shown in A.
In order to accurately estimate area, a device should be laid out according to its
maximum density rules, and its resistance extracted. This produces a better estimate for
the specific on resistance. In this case it was found that the value was about 20% higher
than the one provided by the foundry.
The input capacitance and output capacitance of the transistor are extracted as well.
These are used to determine Coss density and Cin density. These three parameters can be
used to determine conduction losses, switching losses, and gate drive losses.
3.2.3 Transistor Sizing
The sizing of each transistor was determined using a convex optimization procedure as
described in [2]. A summary of this process is described below.
A fixed area amount of 3.5 mm2 was used. This area was based on the amount of area
which would be available after other components including esd protection, the padframe,
and the gate drivers were included.
The script initially sizes each device equally. The script than optimizes efficiency for
a single operating point while also ensuring the efficiency remains above certain other
Chapter 3. IC Implementation of Emerging Topologies 29
targets for other operating points. The parameters extracted previously are used to
determine conduction losses, Coss losses and gate drive losses.
It iteratively runs simulations for that point and adjusts device sizings until the
solution converges on the optimal distribution of silicon area for that operating point.
As the converter is targeted for wide input and output range operating conditions, several
points of optimization were selected, and simulations were run in Cadence to determine
which was the most desirable across all operating points. The design ultimately selected
was the one optimized for the 24 V to 5 V operating point as this was able to produce
satisfactorily high efficiencies for high-step down while still performing well for conversion
ratios closer to one. A simplified testbench to accurately capture the losses while being
relatively quick to run was constructed. As this testbench simulates real devices, the
technology parameters are automatically modeled. This includes conduction, switching,
Coss losses, and gate drive losses. The DC resistance of the inductor is taken from
the datasheet. As no SPICE model was available for the selected inductor, a simple
ripple-dependent model was used by taking core loss values from Wurth Elektronik’s
REDEXPERT tool for different ripple values and fitting them to a curve. These losses
were then included in the efficiency measurement. The final device sizes are shown in
3.3. Parameters for the seven transistors are provided in table 3.1. Both pre-layout and
post-layout parameters are included. The post-layout resistance includes the resistance
added by the metal to route to the next device. As can be seen this can increase the
resistance of some devices by up to 25%.
Figure 3.3: Pie chart showing area of transistors
Chapter 3. IC Implementation of Emerging Topologies 30
Table 3.1: Device Parameters
Pre-Layout Post-LayoutRon (mΩ) Cin (pF) Coss (pF) Ron (mΩ) Cin (pF) Coss (pF)
Q1 79.8 292 141 95.7 327 165Q2 55.9 417 174 75.85 468 229Q3 51.44 452 188 67.5 512 251Q4 80.6 289 120 99.6 324 159Q5 76.6 303 127 101.6 344 128Q6 78.8 295 125 91.1 330 165Q7 104.3 224 92 131 250.5 125
3.3 Floorplanning
An integrated power stage will have numerous connections both between silicon devices
on-chip and between these devices and external components. The complexity of the
topology requires that special attention be paid to its floorplan. Failing to floorplan
properly can result in increased routing resistance and poorer efficiencies. In this section,
a strategy for proper floorplanning of integrated power stages will be described. The
following principles were developed to describe what a suitable floorplan would look like
and how to achieve it.
1. Place devices in locations which are pad accessible,
2. Arrange devices to minimize routing resistance and inductance between devices,
3. Permit peak current requirements be met from an electromigration perspective,
4. Whenever possible, arrange devices in such a manner which simplifies the layout of
the PCB both in terms of the power stage and connecting passive elements.
In order to create a good floorplan, it is important to consider the highest average
currents in all branches. In Fig. 3.4 the highest average currents are shown in graph
form. The vertices of the graph being pads and switches, and the edges connecting them
being conduction paths. Using this figure a desirable floorplan becomes clearer. Devices
should be arranged in such a way that they can conduct high current to both related
devices and pads. The following principles were developed to migrate from this graph to
a proper floorplan.
Chapter 3. IC Implementation of Emerging Topologies 31
Figure 3.4: Graph showing peak currents on interconnect between transistors and pads
1. Devices should share their perimeters with switches and pads they connect to,
2. Devices which have high current paths to the same pads should be both adjacent
and share an edge of the die. This ensures both devices have access to the pads, re-
duces the number of pads requires and helps reduce losses by sharing metal between
the two devices,
3. The shared perimeter of devices which conduct current with each other, as well as
that between pads and devices conducting with them, should be maximized.
In accordance with the principles above, it becomes clear that all devices must share
their perimeters with the pad-ring as all have high-current paths to one or more pads.
Furthermore, to maximize efficiency and simplify the floorplan as much as possible, de-
vices which conduct high-current to multiple separate pads are placed at the corners of
the die. This means two edges of each of these devices have contact with the padframe,
maximizing the space which can be used to route out to the padframe.
Chapter 3. IC Implementation of Emerging Topologies 32
From this point there are several distinct floorplans which could be used. The one
shown in Fig. 3.5 was selected as it seemed to be the best fit out of the possible options
from a PCB-layout perspective. It allows both inductors to be placed adjacent to one
side of the chip simplifying the path to the output, and preserves small current loops for
the input and flying capacitors. It also offered reduced metal resistance for Q5 and Q7
as it allowed them to share metallization. In addition to this the pads connecting to the
inductors, flying capacitors, and input voltage capacitors are all on different edges of the
chip, preventing the crossing of high current paths, simplifying the PCB layout. This
also permits the various switching nodes and high didt
loops in the main power path, to
be placed on a single-layer of the PCB as shown in Fig. 3.5b. This is beneficial as it is
important to minimize the loop inductance of these paths, and generally ill-advised to
via to multiple layers.
Figure 3.5: A: Floorplan, shows the placement of the switches on the die B: Shows howcritical loops may be placed on a PCB
Chapter 3. IC Implementation of Emerging Topologies 33
3.4 Top Metal Utilization
In the available technology six metal layers were used. The top metal layer has very high
current carrying capacity, exceeding that of all other layers combined, as such it should
be distributed with special attention. This is illustrated in Fig. 3.6. Fig. 3.7 shows the
distribution of the top metal layer in the layout. The colors on the figure represent which
net corresponds to which on the schematic. It is also important to remember the metal
is not placed as one sheet covering the entire indicated area. The indicated areas simply
show which net the usage of the top metal is reserved for in a given area.
Figure 3.6: Shows the metal thickness of the technology utilized. The top metal layer is8x thicker than the low metal layer, and can handle 8x more current. It is rated for 8mA/μm
Due to the complex nature of the topology the lower metal layers are also utilized
extensively. They are most frequently utilized to distribute current laterally across an
individual device, but are sometimes also used to conduct current to other devices when
it is too problematic to do so using the top metal layers. The brown bars in 3.7 indicate
where lower metal layers are used to route between devices (or between devices and pads).
This was done between Q6 and ground, Q2 and ground, and between Q4 and both Q5/7.
Chapter 3. IC Implementation of Emerging Topologies 34
Figure 3.7: A: Shows the area where top metal usage is reserved for specific nets. Forinstance, the area colored red is where top metal usage is assigned to the net labelledIND2. The brown bars indicate where nets pads are connected using lower metal layersexclusively. B: shows the corresponding color coded nets in the schematic.
For most devices, top metal is reserved for the drain at the input of the device and for
the source at the output of the device. The proportion for each gradually shifts from the
drain to the source as the device moves from its input to its output. Lower metals are
used to pass current in cases where the shared perimeter is very broad, and the direction
of current flow makes it inconvenient to share the top metal between the drain and the
Chapter 3. IC Implementation of Emerging Topologies 35
source.
3.5 Gate Driving
3.5.1 Low Side Gate Driver
The low side gate driver, pictured in Fig. 3.8 consists of a low-voltage (0-5V) level
shifter and an inverter chain. On chip the high current power ground and low current
signal ground are completely separated. Signals which come on chip are referenced to
the signal ground. As there may be a small dc offset between the signal ground and the
power ground a level shifter is needed to ensure safe translation, a standard CMOS level
shifter is used for this purpose.
Figure 3.8: Block diagram showing schematic of low-side gate driver
The driver chain consists of a series of CMOS buffers using a fanout of four. A fanout
of four was used as this is the commonly quoted optimal number to minimize delay
[22]. The sizing of the gate driver has several consequences. A gate driver that is too
weak increases switching losses by reducing the switching time of the power transistor.
However, oversizing the gate driver has several drawbacks as well.
1. Greater area consumption,
2. Can cause EMI related problems and worsen ringing,
3. High current spikes can increase the risk of latch-up.
Thus, depending on area constraints, it is desirable to make the gate-driver just slightly
stronger than it needs to be. This is because the strength of the driver can easily be re-
duced after the fact by adding a supply/bootstrap resistor to limit the maximum amount
of current which can be pulled into the driver. Table 3.2 shows losses for different gate-
source rise times, for a high-side switch operating for a 24 V input voltage and 2 A
load current. As can be seen, for the selected frequency, the losses cease to increase
Chapter 3. IC Implementation of Emerging Topologies 36
significantly for rise times of less than 13 nanoseconds, hence a rise time less than this
value is desirable. Some margin is provided to account for increases in resistance due
to parasitics, so the inverter chain was designed for a 10%-90% rise time of 10ns, for
the largest device on the chip Q3. Doing so also allows the frequency the prototype’s
switching frequency to be increased if desired without incurring unduly high switching
losses.
Table 3.2: Driver Chain Strength
Rise time (ns) SWloss (mW)
42 6636 63.130 60.520 5713 53.456 52.89
3.5.2 High Side Gate Driver
The high side gate driver is more sophisticated than the low-side one. High-side gate
drivers are used to drive so called high-side devices. These are used to drive NMOS
devices whose source nodes are not necessarily tied to ground. The challenges of this
can best be explained referring to Fig. 3.9 This figure shows a simple half-bridge. For
the technology the transistors are made out of, the maximum permissible gate-source
voltage is 5 V. This can be understood by looking at a buck converter for the case when
the converter is transitioning from the state where its low-side device Q2 is on and its
high-side device Q1 is off to the state where Q1 is on and Q2 is off. When Q1 turns
on, its source is initially zero volts. When the device is on, the source should rise to 12
V. This requires the gate voltage to be 17 V, in order for the transistor to be fully-on.
This is accomplished by using a 5 V supply referenced to the source node of the the
transistor, that is, the gate driver supply voltage is always 5 V higher than the source
of the transistor. When the device is to be shut-off, the gate driver outputs the source
voltage. Q1,3,4,5,7 all require high-side gate drivers.
The scheme for powering the high-side gate driver shown in 3.9 is referred to as
bootstrapping. It is used when the source of a n-type power device is switching between
ground and another potential. A supply charges the bootstrap capacitor while the source
is pulled to ground. When the source rises above ground, the bootstrap voltage rises with
Chapter 3. IC Implementation of Emerging Topologies 37
it, and the diode becomes reverse biased. More details of the bootstrapping scheme used
for the 7SFC are included in Appendix A, which was previously described in [11].
Figure 3.9
A block diagram of this driver is shown in Fig 3.10. There are three basic components,
a level-shifter, a schmitt trigger, and a driver chain, which is identical to the driver chain
used in the low-side gate driver. Unlike the low-side level shifter, which is only shifting
the signal with respect to its logic-low supply, the high-side shifter is shifting the signal
with respect to both its logic-low and logic-high supply. The logic-low voltage is shifted
from ground, to a voltage VS, which may be anywhere from 0 V to 42 V. The logic-
high supply is being shifted from 5 V, to a value VB, which is equal to VS + 5V . The
output of the shifter must stay between VS and VB, otherwise it may damage stages
downstream of the level-shifter (and possibly the shifter itself) by exposing transistors to
overvoltage conditions. This is because the maximum permissible gate-source voltage is
5 V regardless of the rating for the drain voltage of the device. It is also desirable that
there be no static-current draw between VB and ground, as this potential difference may
be as high as 47 V and thus result in high power consumption.
Chapter 3. IC Implementation of Emerging Topologies 38
Figure 3.10: Showing function of high side driver, when on, output must remain 5 Vabove source voltage
The selected level shifter is the one described in [24]. The schematic is shown in Fig
3.11. This shifter uses a differential pair of transistors which has no static tail current.
When the input of Q1 goes high, it pulls the voltage VOut low. Transistor Q3 prevents
this voltage from going below VS +Vthreshold, where Vthreshold is the threshold value of the
PMOS device. This prevents the output from dropping below VS and exposing devices
downstream of the shifter to an overvoltage condition. This branch going low causes Q8
to turn on and pull VOutB high. This turns Q7 off and Q5 on. This causes Vout to be
pulled all the way down to VS. As Q7 is off, no current will run through the circuit
at this point. When the input is low, the reverse happens, where VOutB goes low and
VOut goes high. The output swing of this level shifter is shown for a number of process
and temperature corners below 3.12. Characteristics of the level-shifter and gate driver
are shown in table 3.3. As current is drawn only when the circuit is being switched,
or when VS is changing, there is no static current draw, which greatly limits the power
consumption. Unlike with competing designs discussed in Appendix D, the inputs are
always active, and will work to keep the circuit in its desired state regardless of whatever
disturbances effect it.
Chapter 3. IC Implementation of Emerging Topologies 39
Figure 3.11: Selected level shifter using cross-coupled topology
Figure 3.12: Process and temperature simulations for the level-shifter operated for Vs/VB
of 0/5 and 48/53
Downstream of the level shifter are the schmitt trigger and the driver chain. The
Chapter 3. IC Implementation of Emerging Topologies 40
schmitt trigger is used to provide hysteresis. This is important as the power supply may
be very noisy, especially as the supply is floating, and not sitting at a single potential
as referenced to the ground used on the chip. Hysteresis helps prevent the possibility of
the shifter from spontaneously switching back during a transition. The driver chain is
identical to the driver chain used in the high-side gate driver.
Table 3.3: Level Shifter Parameters
Parameters Min Max
Power (μW) 46 360Rise Time (ns) 6 12
Propagation Delay - Gate Driver (ns) 9 34
3.5.3 Gate Driver Layout
The layout of high side driver is shown below. As this is a circuit which will draw very
large amounts of current very suddenly, and contains both n and p devices, special care
was taken during its layout. The n and p devices were placed in separate “deep trenches.”
These trenches consist of large insulators inserted into the substrate to separate it from
the surrounding area. Although this increases the amount of area consumed, it helps
prevent latch-up, where current injected into the substrate causes a positive feedback
loop to form between parasitic devices in the circuit eventually resulting a low-impedance
path directly between the supply voltage and ground.
Figure 3.13: Layout of gate driver
Chapter 3. IC Implementation of Emerging Topologies 41
3.6 Top-Level Layout
The top level layout of the chip is shown below in Fig. 3.14.
Figure 3.14: Top Level layout of 7SFC including gate drivers and transistors
Figure 3.15: Switches and gate drivers boxed in using black, pads shown in red
Chapter 3. IC Implementation of Emerging Topologies 42
Table 3.4: Inductor Parameters
Manufacturer Part Number Inductance DCR Package Name Volume
Wurth 74479299222 2.2μH 106 mΩ PMCI 1210 9.6mm3
Wurth 74479276222C 2.2μH 135 mΩ PMCI 0806 LR 3.52mm3
In Fig. 3.15 the power transistors, gate drivers, and high-current pads are labeled.
The devices are arranged in the manner of Figs. 3.5 and 3.7.
3.7 Experimental Results
The IC was fabricated in a 0.18 μm 80 V technology (using 24 V devices). Fig.3.16 shows
a photo of the die taken after receiving it back from the foundry.
Figure 3.16: Photo of chip in package with lid removed
The chip was placed in a 48 pin QFN package. A PCB was developed to test this chip.
Measurements were made on the chip using two different sets of inductors. Properties
for these inductors are listed in Table 3.4. The volume provided refers to the volume of a
single inductor of that class, hence the total inductor volume used by 7SFC is twice that
value. Listed in 3.5 are additional components that were used for the 7SFC. There are
the bootstrap diodes, bootstrap capacitors, flying capacitor, and output capacitor. The
Chapter 3. IC Implementation of Emerging Topologies 43
Table 3.5: Other Important Components
Part # Used Part Number Parameter Volume
Bootstrap Diode 5 CUS05S40 40 v 4.1mm3
Bootstrap Capacitor 5 GRM155R62A104KE14 100 nF 0.275mm3
Flying Capacitor 2 C2012X5R1H475K125AB 4.7μF 1.1mm3
Output Capacitor 1 C1210C476M4PACTU 47μF 22.4mm3
bootstrap diodes can be integrated on-chip, but it was decided for this prototype to use
off-chip diodes to permit more flexibility during testing. The bootstrap capacitors cannot
be easily integrated on-chip and thus must be used as an external components. The total
volume of 5 bootstrap capacitors is 1.375 mm3, which is insignificant as compared to
other parts of the converter.
3.7.1 Switching Waveforms
The switching nodes and inductor currents for the three major operating modes are
shown below. Additional waveforms showing the gate voltages are provided in Appendix
B. Fig. 3.17 shows the HSD Buck operating region for an input voltage of 12 V. As can
be seen the swiching node voltages go between 0 and 6 V. Fig. 3.18 shows waveforms
for the 3-level buck. The two inductor current waveforms are virtually identical for this
mode, so they are shown with different offsets. The switching waveform is about 6 V as
well. Fig 3.19 shows waveforms for the interleaved buck for an input voltage of 5 V. As
can be seen the switching node voltages are 5 V each as expected. This indicates the IC
is working as expected. Extensive tests were run for a wide range of operating points, the
switching nodes for 42 V operation is shown for the HSD operating mode in Appendix
B as well.
Chapter 3. IC Implementation of Emerging Topologies 44
Figure 3.17: Waveforms for 7SFC operating as a HSD Buck, showing inductor currentsand switching node voltages. Inductor currents measured with same offset to show inter-leaved nature and equal current sharing. Ch1: Switching node 1 voltage, 5 V/div. Ch2:Switching node 2 voltage, 5 V/div. Ch3: Current of L1, 500 mA/div. Ch4: Current ofL2, 500 mA/div. The time scale is 500 ns/div.
Figure 3.18: Waveforms for 7SFC operating as 3-Level Buck, showing inductor currentsand switching node voltage,as inductor waveforms are nearly identical different offsetwas used to show difference clearly. Ch1: Switching node 1 voltage, 10 V/div. Ch2:Switching node 2 voltage, 10 V/div. Ch3: Current of L1, 500 mA/div. Ch4: Current ofL2, 500 mA/div. The time scale is 500 ns/div
Chapter 3. IC Implementation of Emerging Topologies 45
Table 3.6: Volume Comparison Between 7SFC and Similar Products
Part Number Inductor Flying Capacitor Packaged Silicon Total
LM25011[25] 450mm3 0 37.17mm3 487.17mm3
LMR24220[26] 251.3mm3 0 3.45mm3 254.75mm3
7SFC 1210 19.2mm3 2.2mm3 30.24mm3 51.64mm3
7SFC 0806 7.04mm3 2.2mm3 30.24 mm3 32.928mm3
Figure 3.19: Waveforms for 7SFC operating as 2-phase interleaved buck, showing inductorcurrents and switching node voltages. Ch1: Switching node 1 voltage, 10 V/div. Ch2:Switching node 2 voltage, 10 V/div. Ch3: Current of L1, 500 mA/div. Ch4: Current ofL2, 500 mA/div. The time scale is 500 ns/div
3.7.2 Conventional Solutions and Passive Volume Comparison
Efficiency results for conventional solutions were provided to us by Wurth Elektronik.
These results were gathered using the evaluation boards provided by the companies. The
volume of these solutions are compared to that of the 7SFC when implemented with
0806 inductors and 1210 inductors in Table 3.6. Table 3.7 compares the surface area
consumed as well. Packaged silicon area refers to the area of silicon devices including
packaging used in the various solutions. For the LM25011 this includes an external diode
also placed on the board (as this IC does not contain a synchronous rectifier). In the
following section, the efficiency of the 7SFC will be compared to that of these solutions.
The volume and area comparisons of Tables 3.6 and 3.7 do not include the bootstrap
Chapter 3. IC Implementation of Emerging Topologies 46
diodes and capacitors. As mentioned before the diodes can be integrated and thus not
a major concern. The bootstrap capacitors generally cannot be integrated on-chip. In
terms of bootstrap capacitor volume, the 7SFC will require about 5 times more volume
than a conventional buck. However, as this value is quite small, this is not a major
concern in and of itself. The presence of these caps does present a drawback that is more
difficult to quantify. They must be placed and routed to, which can lead to consuming a
larger amount of pcb area then they nominally do. They also require more pads be added
to the IC, which is critical in certain applications. Hence, despite being small overall,
they still detract from the merits of the 7SFC.
Table 3.7: PCB Area Comparison Between 7SFC and Similar Products
Part Number Inductor Flying Capacitor Packaged Silicon Total
LM25011 100mm2 0 28.79mm2 128.792
LMR24220 57.76mm2 0 8.64mm2 66.4mm2
7SFC 1210 19.2 mm2 4.8mm2 36 mm2 56.8 mm2
7SFC 0806 5 mm2 4.8mm2 36mm2 45.8 mm2
3.7.3 Efficiency Results and comparison
Efficiency results for the 7SFC are presented below. The “target” curve refers to the
best efficiency measured from the products the 7SFC is compared to. That is, mea-
surements were taken from a number of existing solutions and the best efficiencies were
compiled into the target curve provided above. Figure 3.20 shows efficiency results us-
ing the PMCI 1210 inductors on the 7SFC where 3.21 shows results using the smaller
PMCI 0806 LR inductors. Generally speaking the 7SFC beats all efficiency targets for
medium to heavy load when using the 1210 inductors. Most targets are met at medium-
heavy load using the smaller 0806 LR inductors. All efficiency results for the 7SFC were
gathered in forced Continuous Conduction Mode (CCM). This is the mode of operation
available in synchronous converters where the inductor current is allowed to go negative.
Discontinuous conduction mode (DCM) on the other hand is where conduction ceases
when the inductor current reaches zero. DCM occurs naturally in bucking topologies
where the low-side devices are implemented as diodes, as these cannot conduct negative
current. In synchronous converters the inductor current must be sensed and the low-side
switch switched off just before the current goes negative in order to achieve DCM. As
neither DCM nor Pulse Frequency Modulation (PFM) were implemented, the light-load
efficiencies listed are pessimistic. PFM is a technique where the frequency is reduced
Chapter 3. IC Implementation of Emerging Topologies 47
while operating at light-load in DCM to reduce switching and gate drive losses while
still meeting ripple specifications. If DCM or PFM are implemented the efficiency could
improve considerably although it is unknown if it will be better than competing products
operating in that mode.
Generally speaking, for the HSD mode, lower input voltages and higher output volt-
ages favor higher frequencies, and vice versa. More precisely, lower input voltages but
higher inductor current ripples favour higher frequencies. This is because higher frequen-
cies reduce RMS current by reducing the ripple, but increases Coss losses, which are made
worse by higher voltages. The other operating modes should also work better for higher
switching frequencies when the inductor current ripple is high and the input voltage is
low.
Chapter 3. IC Implementation of Emerging Topologies 48
Figure 3.20: Efficiency curves for the 7SFC using 2.2μH inductors in a 1210 package.Title indicates input and output voltages as well as operating mode (3LB = 3 LevelBuck, HSD = High Step Down Buck). Frequencies indicated in legend.
Chapter 3. IC Implementation of Emerging Topologies 49
Figure 3.21: Efficiency curves for the 7SFC using the 2.2μH inductors in an 0806 LRpackage. Title indicates input and output voltages as well as operating mode (3LB = 3Level Buck, HSD = High Step Down Buck). Frequencies indicated in legend.
Chapter 4
Single-Mode Controller with Near
Minimum Deviation Transient
Response
Modern switch-mode power supplies (SMPS), especially those used in point-of-load (PoL)
applications, need to meet challenging requirements for both steady state and dynamic
voltage regulation [27], [28]. In targeted applications, for converters processing from
a fraction of a watt to tens of watts, the dynamic performance of SMPS controllers
have a dominant influence on the sizing of the output filter capacitor [20]. As required
output voltages decrease and load currents increase, meeting output voltage regulation
requirements during transients usually requires greatly increased output capacitance
[20],[29],[30],[31]. Power supplies in portable applications have already been shown to
consume a significant portion of PCB real estate [20]. Supply voltages for processors and
other integrated circuits have been continuously decreasing. As output voltages decrease,
and load currents increase, the required size of output filter capacitors continues to grow
[30]. Therefore, it is important to improve the dynamic performance of controllers as
much as possible to limit this growth in area consumption.
This chapter introduces the practical single-mode minimum-deviation concept based
controller of Fig. 4.1. Unlike previous solutions, it uses only one mode of operation for
both steady state operation and during transients, eliminating potential mode transition
related stability problems while also further simplifying the hardware implementation
requirements.
50
Chapter 4. Single-Mode Near Minimum Deviation Controller 51
Figure 4.1: Block diagram of the proposed controller used in a single-phase buck converter
4.1 Review of Fast Transient Controllers
As discussed in section two many fast-transient controllers have been developed. Two
kinds are minimum deviation controllers and time-optimal controllers. Time-optimal
controllers aim to recover voltage in the fastest amount of time possible. They do this by
calculating the current over/undershoot required to recover lost charge within a single
cycle and then returning the inductor current to its steady-state value. However, they
suffer from high computational complexity and require the usage of power hungry high-
speed analog to digital converters [20]. Furthermore they are very sensitive to non-ideal
behavior in converters (losses and parasitics)and small variations in the inductor used
can greatly impact the transient response and lead to voltage overshoots and undershoots
[20]. This in turn can cause the converter to toggle back and forth between its transient
controller and steady-state controller for a single transient.
Several minimum deviation controllers have been proposed. These controllers respond
to load transients in the fastest possible time. They do not suffer from the computational
Chapter 4. Single-Mode Near Minimum Deviation Controller 52
complexity of time-optimal controllers and have reduced requirements for analog hard-
ware [20]. Furthermore they do not require accurate knowledge of converter parameters
and are mostly insensitive to losses and parasitics. They are however still sensitive to
delays which can result in current over/undershoots. This in turn can lead to toggling
problems. Partly as a result of concerns such as these, control schemes such as these
have not been widely adopted.
This chapter will introduce a single-mode minimum deviation controller. This con-
troller is not susceptible to the toggling issues present in two-mode solutions and is
simpler to implement due to lack of mode-changing and transient detection circuitry.
4.2 Principle Of Operation
The introduced controller is shown in Fig. 4.1. It is a modification of a voltage mode
controller where the conventional pulse width modulator (PWM) is replaced by the load
tracking modulator block. This block consists of two half duty digital pulse width modu-
lator blocks (DPWM1 and DPWM2) and one capacitor current polarity detection block.
The half-duty modulators use information about the capacitor current polarity to gate
the power transistors (Q1 and Q2) such that the inductor current is made to match the
load current cycle by cycle while also reconstructing the desired inductor current ripple.
It does this by splitting both the on-time (Q1 on, Q2 off, inductor charging) and the
off-time (Q1 off, Q2 on, inductor discharging) into two separate phases each. These are
called the current correction phase (CCP) and ripple reconstruction phase (RRP).
The functionality of the controller can be better understood with the help of Figs. 4.2
and 4.3. Fig. 4.2 depicts the operation of a buck converter running with the introduced
control scheme in steady-state. At point 0, the converter enters its on-state. Q1 is on, Q2
is off, and the inductor current is charging up. This first segment is its on-state current
correction phase (CCP). This phase runs until the inductor current equals (or otherwise
exceeds) the load current (point 1 in Fig. 4.2), which in turn causes the capacitor current
polarity to change and become positive. This sends the output of the polarity detector,
vvpd(t) high. This sets the half-duty modulator DPWM1 which marks the beginning of
the on-time ripple reconstruction phase (RRP). Once set, DPWM1 keeps the converter
in its on-state for an additional DTs nom/2 in order to reconstruct its ripple. At this
point the output of DPWM1 will go high and push the converter into its off-state. This
is marked by point 2 in Fig. 4.2. Now, Q1 is turned off, Q2 is turned on and the inductor
current is discharging. The first portion of its off-state is again a current correction
Chapter 4. Single-Mode Near Minimum Deviation Controller 53
phase. It will remain in its current correction phase until the instantaneous inductor
current equals (or otherwise is less than) the load current. Which occurs at point 3.
The capacitor current again experiences a change in polarity, setting vpd(t) low. The
converter is again in a ripple reconstruction phase. This activates DPWM2 which keeps
the converter in its off-state for an additional (1-D)Ts nom/2 to reconstruct the ripple.
This is point 4 in Fig. 4.2. The output of DPWM2 then goes high, putting the converter
back into its on-state, and the cycle repeats.
Figure 4.2: Ideal waveforms of the proposed control scheme operating in steady-state.From top to bottom are the output voltage and nominal output voltage on the sameaxis, the inductor current and load current on the same axis, the capacitor current, theoutput of the polarity detector, gating signal of Q1, and gating signal of Q2.
In the event of a transient, the controller still operates just as it does in steady-state.
However, the load-tracking modulator will extend either the on-time or the off-time
(depending on the transient) and thus the period for that cycle to address the transient.
This is shown in Fig. 4.3 for a light-to-heavy transient. As the output of the polarity
detector cannot go high until the inductor current equals (or exceeds) the load current,
Chapter 4. Single-Mode Near Minimum Deviation Controller 54
the duration of the CCP of the on-state is automatically extended. Once the inductor
current has increased to the point that it equals the load current, the converter enters
its RRP and the controller proceeds as before. Heavy to light transients are addressed
similarly, however it is the CCP of the off-state that is extended as opposed to that of
the on-state. From this it can be seen that this control method requires no special blocks
to detect load transients, it simply responds to them as part of its normal operation.
Chapter 4. Single-Mode Near Minimum Deviation Controller 55
Figure 4.3: Ideal waveforms of the proposed control scheme operating during a light-to-heavy load step. From top to bottom are the output voltage and nominal output voltageon the same axis, the inductor current and load current on the same axis, the capacitorcurrent, the output of the polarity detector, gating signal of Q1, and gating signal of Q2
As shown in Fig. 4.3, there will be an error in the voltage even after the inductor
current has been corrected. This happens because some charge is lost before the inductor
current is corrected. Voltage recovery is handled by the compensator afterwards as
shown in Fig. 4.4, this is similar to conventional voltage mode control. The speed of this
recovery depends on the compensator implemented. This controller may be thought of as
Chapter 4. Single-Mode Near Minimum Deviation Controller 56
a voltage mode controller with one major difference, it is capable of correcting inductor
current within a single cycle thus achieving near minimum output voltage deviation.
Experimental results taken from a physical prototype are presented in the final section.
Figure 4.4: Voltage recovery after load transient for single-mode controller
4.2.1 Analysis of Switching Period Variation
As the gates are not modulated by fixed-frequency signal, the period/frequency can
vary slightly with perturbations. This section will describe the extent to which this can
happen by. Analysis will also be presented to show that in the absence of perturbations
the period will be equal to the nominal period described above.
For the converter’s on-time, the duration of the current correction phase can be
determined using the following Equation
TON = DTsnom +∆iLtransVg−Vout
L
(4.1)
Where ΔiLtrans is the magnitude of any load step that may have occurred. During steady-
state operation ΔiLtrans=0, thus the converter’s on-time which is equal to the sum of its
CCP and RRP during steady-state operation is found by the following equation
TON SS =DTsnom
2+DTsnom
2= DTsnom (4.2)
Similarly, in the absence of load-steps, the CCP and RRP of the off-time have equal
Chapter 4. Single-Mode Near Minimum Deviation Controller 57
durations of (1 −D)Tsnom/2. The sum of these provide the off-time.
TOff SS =DTsnom
2+DTsnom
2= DTsnom (4.3)
Hence the effective steady-state period TSW SS defined as the duration required to com-
plete steps 0 through 4 in the absence of disturbances, will be equal to Ts nom.
TSW SS = (1 −D)Tsnom +DTsnom = Tsnom (4.4)
In the case of a load change between points 0 and 1, shown in Fig. 4.2, the switching
period of the converter changes over one switching cycle. In this case, the switching
period can be, calculated using the same procedure and taking the non-zero value of
iL trans into account, i.e.
TSW trans = (1 −D)Tsnom +DTsnom +∆iLtransVg−Vout
L
(4.5)
Similar calculations can be made for a heavy-to-light transient.
TSW trans = (1 −D)Tsnom +DTsnom +∆iLtrans
Vout
L
(4.6)
In general the period of the converter will be extended only when a transient occurs.
Otherwise, it will be equal to Tsnom.
4.2.2 Stability for Small Perturbations
Certain control schemes, like peak current mode control can be made unstable by small
perturbations in certain operating regions. For a buck converter this happens for con-
version ratios greater than 0.5. This happens because only the on-time of the converter
is related to a physical measurement (the peak current). Whereas the off-time consists
simply of the remaining time in the switching period. If the flux-linkage across the induc-
tor is larger during the off-time, than the on-time, errors in the duration of the on-time
get passed to the off-time, causing the error in the inductor current to increase leading
to bifurcation. The proposed control scheme does not suffer from such issues as both the
on-time and the off-time are related to the point where a polarity change in the capacitor
current is detected. Hence small perturbations are rejected within a half a cycle.
Chapter 4. Single-Mode Near Minimum Deviation Controller 58
4.2.3 Extension to Multi-Phase Operation
This control scheme can be extended to multi-phase converters (such as the 2-phase
interleaved buck. This can be accomplished by comparing the inductor current in each
phase to half of the load current as opposed to using the estimated capacitor current. Fig.
4.5 shows key waveforms for the two-phase case where the period is 2Ts. The modulator
behaves much the same as it does in the single-phase case, however instead of using the
zero crossing point of the capacitor current (which corresponds to a shift in polarity) it
uses the point where the inductor current equals half the load current. Furthermore, the
second phase may not enter its on-state until Ts has elapsed after the first phase entered
its on-state. Following this approach this control scheme may be extended to n-phases
by comparing the current in each phase to the load current divided by n. An added
benefit of this control scheme is that it ensures currents are shared equally in multi-
phase converters, as it forces the average current of each phase to equal iLoad(t)/n. This
eliminates the need for a current-balancing algorithm, simplifying the system. However,
unlike the single-phase implementation, it might not be possible to implement with a
single non-intrusive sensor.
Figure 4.5: Showing response of controller in multi-phase case
Chapter 4. Single-Mode Near Minimum Deviation Controller 59
4.2.4 Extension to Indirect Energy Transfer Converters
Figure 4.6: Controller modified to be used in a boost converter, note modulator remainssame, only difference is capacitor time constant estimator replaced with 2 sensors anddigital potentiometer
Figure 4.7: Showing waveforms for control scheme applied in boost converter
Chapter 4. Single-Mode Near Minimum Deviation Controller 60
This control scheme can theoretically be extended to converters employing indirect energy
transfer, examples being a boost or buck-boost converter. In such converters, the inductor
current does not equal the load current, however it is proportional to it. Hence it may
be possible to apply this scheme by comparing the scaled inductor current to the load
current (or vice versa). A schematic depicting how this would look is shown in Fig 4.7.
This scheme takes advantage of the following expression.
iL(t) = iLoad(t)/(1 −D)
Hence, whereas in a buck converter, the inductor current will be equal to the load
current mid-cycle, in the boost, the inductor current will equal the load current, divided
by (1 − D) at this point. Using this to trigger the half-duty modulators instead of the
zero-crossing point of the capacitor current allows the controller to be implemented in
the same way. There are many possible ways to implement it. Fig 4.7 is one possible
implementation. Where 1 − D is passed to a digital potentiometer, which divides the
inductor by 1/(1−D). This is then passed to a comparator, which compares it with the
load current. the output of this is used to trigger the half-duty modulators in the same
way the polarity detector did previously.
It is important to note however this relationship is not perfectly accurate as all con-
verters are lossy. In very lossy converters the difference between this expression and the
true one accounting for losses can be dramatic and could possibly complicate our prevent
the successful implementation of such a controller.
4.3 Practical Implementation
This section will describe implementation details for this controller as applied to a conven-
tional single-phase buck converter. Discussed is the structure of the half-duty modulator
and the implementation of capacitor current estimation techniques.
This control scheme makes use of the capacitor current. More specifically, this con-
troller requires the accurate detection of changes in the polarity of the output capacitor
current. To avoid placing an intrusive sensor in series with the capacitor an estimation
circuit is used. This is done by placing an RC circuit in parallel with the output ca-
pacitor whose time constant is matched with that of the output capacitor and its ESR
[32],[33],[34]. This is shown in Fig. 4.8. Previous implementations have demonstrated
that auto-matching implementations of this approach can accurately detect the capacitor
current polarity and zero current crossing points with a low silicon area [16].
Chapter 4. Single-Mode Near Minimum Deviation Controller 61
Figure 4.8: Structure of the RC time constant estimator and zero-crossing detectioncircuit
Fig. 4.9 depicts a block diagram of the half-duty DPWM cell. It consists of a counter
and a digital comparator. The set pin is triggered by a pulse of any duration. It causes
the counter to begin counting up. The output of the half-duty DPWM cell goes high
once the counter exceeds the value of the reference supplied to it. The reset pin can also
be triggered by a pulse of any duration, and causes the counter to reset to zero. After a
reset, the counter will not count until another set pulse has been received. The counter
may only be set if the output of the DPWM is low and only reset if the output of the
DPWM is high.
Figure 4.9: Block diagram of half-duty modulator
Chapter 4. Single-Mode Near Minimum Deviation Controller 62
4.4 Experimental and Simulation Results
To verify the effectiveness of the introduced method an experimental prototype was made
based on the diagrams of Figs. 4.1, 4.8 and 4.9. A single-phase buck converter based
prototype was made with a switching frequency of 600 kHz. Several properties of this
prototype are shown in Table 4.1. The digital controller is implemented using an FPGA
based system, while the additional analog components, shown in Figs. 4.1 and 4.8 are
implemented using off-the-shelf ADCs, comparators and operational amplifiers. Fig. 4.10
depicts simulation results whereas Figs. 4.11 and 4.12 are physical measurements. Fig.
4.10 shows a simulation of the controller responding to a 4 A , 15% to 100% light to heavy
load step. Fig. 4.11a and 4.11b shows experimental results taken from the prototype
for the same transient, while Fig. 4.11c and 4.11d shows its response to an opposite
transient. As in Fig. 4.10, it can be seen that after the current has been corrected, the
voltage ceases to change. This indicates that the current has been correctly adjusted.
It can be seen in both Figs. 4.11c and 4.11d that the inductor current does not change
significantly after the initial correction. The compensator then works to recover the
voltage difference.
Table 4.1: Parameters of fast transient prototype
Parameter Value
Frequency 600 kHzVg 12 V
Cout 100 μFVout 3.3 VPout 15 WResr 10mΩL 3.3 μH
Chapter 4. Single-Mode Near Minimum Deviation Controller 63
Figure 4.10: Simulation of controller for 4 A load step, depicting output voltage, inductorcurrent, and load current.
Figure 4.11: A: Oscilloscope screenshot showing output capacitor voltage and inductorcurrent after 4 A light to heavy load step, (ILoad 0.6 to 4.6A). Ch1 vout(t) , 100 mV/div.Ch2 iL(t), 2 A/div. Timescale is 2 us/div. B: Zoomed out version of A showing voltagerecovery phase, timescale set to 500 us/div. C: Identical loadstep, except going fromheavy to light, D: zoomed out loadstep shown in C
Chapter 4. Single-Mode Near Minimum Deviation Controller 64
Figure 4.12: A: Controller responding to consecutive load steps. Ch1 vout(t), 100 mV/div.Ch2 iL(t), 1 A/div, timescale is 5 μs B: Reference step indicating controller’s stability.Ch1 vout(t), 1 V/div, timescale is 1 ms
Fig. 4.12a shows the response of the controller for consecutive load transients. After
the first 1.2 A transient occurs the voltage begins to drop, the drop ceases occurring
once the current has been corrected. A second 1.6 A transient then occurs which causes
the voltage to again drop until the current is corrected. Fig. 4.12b shows the output
voltage during a reference voltage step. This transition occurs smoothly indicating the
controllers stability.
It can be seen that the single mode controller reconstructs the inductor current to its
new steady state value over a single switching cycle with near minimum output voltage
deviation. Also, it can be seen that due to the single mode of operation a smooth settling
of the voltage is achieved without toggling problems often existing in dual mode solutions.
Fig. 4.13a shows simulation results for a load-step on a 2-phase converter utilizing
the introduced control scheme. In this simulation the control scheme is implemented
as described previously. It can be seen that the sum of the inductor currents are made
to match the load in just one switching action (per phase) while otherwise causing the
converter to behave like a conventional 2-phase interleaved buck converter.
Fig 4.13b shows simultion results for a boost converter controlled in the manner
described previously responding to a load-step as well. Plotted in the simulation are the
output voltage, load current, and scaled inductor current (as would be at the output of
a digital potentiometer). As can be seen the scaled current is made to match the load
current cycle by cycle.
Chapter 4. Single-Mode Near Minimum Deviation Controller 65
Figure 4.13: A: Depicts a load-step for the control scheme applied to a 2-phase converter.B: Depicts a load-step for the control scheme applied to a Boost converter, showing theoutput voltage and scaled inductor current such as would be seen at the output of thedigital potentiometer.
Chapter 5
Conclusions and Future Work
5.1 Integrated Implementation of Emerging Topolo-
gies
An integrated multi-level converter of significant complexity was designed and imple-
mented. It was designed for wide-input and output operating range conditions. The
on-chip power stage consisted of seven power transistors and seven gate drivers to drive
them.
The implementation of such converters requires careful floorplanning and layout.
Carefully planning the arrangement of devices, using the help of a graph detailing peak
average currents on interconnecting branches was shown to be useful in this regard. Us-
ing such techniques, the added routing resistance could be made a small fraction of the
nominal resistance of the device in most cases, despite using only a single thick metal
layer. High voltage level shifters were shown to be implementable while consuming only
a small fraction of the nominal output power of the IC. This is done by using a topology
that consumes no static current. The size of the transistors used were optimized for good
efficiency across the operating range. The efficiency of the on-chip implementation was
compared to the best achievable efficiency of several competing products on the market.
For most operating points it was superior despite using a much smaller total volume.
This indicates the optimization was successful.
5.1.1 Future Work
The chip remains to be tested in discontinuous conduction mode. Doing so will most
likely improve the efficiency of the converter in light load and may allow it to reach
66
Chapter 5. Conclusions and Future Work 67
performance targets it is currently shy of. Integrating a controller on-board the chip as
well in addition to the power-stage would complete the picture.
5.2 A Single-Mode Load-Tracking Voltage Mode Con-
troller with Near Minimum Deviation Transient
Response
A practical single-mode quasi constant-frequency controller which provides near mini-
mum deviation transient-response has been presented. The introduced controller provides
recovery of the inductor current to its new steady state value within one-switching ac-
tion while eliminating problems related to mode transition existing in previous solutions.
The key new element is the load-tracking modulator, which replaces the conventional
PWM used in normal voltage mode controllers. This modulator forces the inductor cur-
rent to match the load current and then reconstructs the desired ripple waveform. An
experimental prototype was constructed demonstrating this control scheme in a single-
phase buck converter operating at 600 kHz. Its response to single and sequential load
steps was shown. Extensions to both multi-phase and indirect energy transfer converters
were proposed. Simulation results for these extensions were presented demonstrating
near-minimum deviation
5.2.1 Future Work
A simple mathematical model for this control scheme is yet to be developed. One chal-
lenge in analyzing it is that the controller does not operate with a fixed period, thus
analysis techniques usually applied to linear converters do not yield useful information.
A proper s-domain model would be useful for optimizing the compensator in the feedback
loop.
Although a prototype was constructed using a single-phase buck converter, physical
prototypes for multi-phase and indirect energy transfer converters have yet to be con-
structed, and only simulation results have been gathered thus far. Furthermore the pre-
cise details about how they would be practically implemented remain to be worked out.
The construction of such prototypes based on the implementations described in chapter
4, would be good final confirmation of the controller’s validity for such converters.
Appendix A
Bootstrapping Scheme
Bootstrapping is used to power the high-side gate drivers. The scheme used is shown
in A.1, this is similar to the one originally proposed previously [11]. This is based on
the “cascaded bootstrapping” scheme initially proposed in [35]. Switches 3, 5, and 7
all use conventional bootstrapping. When there source voltages drop down to zero, the
bootstrap supply charges to 5 V. When the power device turns on, the source node rises
to the intended voltage, with the bootstrap voltage remaining 5 V above it. Switches 1
and 4 cannot use a scheme this simple as for both switches, there exist modes of operation
where the source node never drops to zero. Hence as opposed to using a 5 V supply to
charge the bootstrapping node, the bootstrap node of switch 5 is used. This can be
explained by example. In the interleaved buck mode of operation, the source voltage of
switches 1 and 4 is always the input voltage. When switch 1 turns on, its source voltage
goes from zero to the input voltage. Its bootstrap voltage then rises to the input voltage
plus 5 V. This charges the bootstrap voltages of switches 1 and 4 to be equal to the input
voltage plus 5 V.
68
Appendix A. Bootstrapping Scheme 69
Figure A.1: The bootstrapping scheme used for power the gate drive circuits in thedesigned power stage of the 7SFC
Appendix B
Additional waveforms for IC
In this section gate drive waveforms are shown for the developed IC operating as an HSD
Buck for an input voltage of 12 V. This is to demonstrate the correct functionality of the
chip. The switching node is also shown for the HSD mode of operation when the input
is 42 V.
Figure B.1: Showing gate waveforms for Q1 and Q2 as well as the two main switchingnodes
70
Appendix B. Additional waveforms for IC 71
Fig B.1 shows waveforms for the gates of Q1 and Q2. When Q1G is low, its voltage is
6 V, equal to Vg/2. When it turns on, it rises to about 17 V, which is equal to the input
voltage plus the drive voltage. When it goes up, vsw1(t) goes up to 6 V, which is equal
to half the input voltage. When Q2 is on, its gate goes to 5 V, and vsw1(t) goes to zero.
Figure B.2: Switching waveforms for HSD mode of operation, showing gate of switch 4and 6, as well as switching node 2
Fig. B.2 shows waveforms for the gates of Q4 and Q6. When Q4 is low, its voltage is
6 V during state 1. This is because Q1 being on, causes the source of Q4 to be equal to
Vg - Vcfly. For states 4 and 2, it is off, causing gate voltage to be zero. During state 3 it
turns on, and gate voltage rises to 11 V, as this increases its source and vfly2 to be 6 V.
When Q6 is on, its gate goes to 5 V, and vsw2(t) goes to zero.
Appendix B. Additional waveforms for IC 72
Figure B.3: Switching waveforms for HSD, showing gate of switch 3, 1, and vsw1(t)
Fig. B.3 shows waveforms for the gates of Q3 and Q1. As Q3 is always on during
HSD mode operation. As a result, Q3g is 5 V when Q1 is off. When Q1 turns on, the
source of Q3 rises to be equal to Vg - Vcfly. This pushes the gate of Q3 up to 11 V.
Appendix B. Additional waveforms for IC 73
Figure B.4: Switching waveforms for HSD, showing gates of switches 7 and 1 as well asvsw1(t)
Fig. B.4 shows the waveforms of the gates of Q7 and Q1. When Q7g goes high, during
state 4, the voltage goes up to 5 V. Once Q1 turns on during state 1, the source and thus
gate node of Q7 rise by 6 V (to be 11 V). When Q1g turns back off, the gate drops down
again to 5 V.
Appendix B. Additional waveforms for IC 74
Figure B.5: Switching waveforms for HSD, showing gate of switch 5, 4, and vsw2(t)
Fig. B.6 shows waveforms for the gates of Q4 and Q5. Q5 has its gate turn to 5 V
during state 2. When state 3 occurs, its source is raised up to 6 V, causing the gate to
rise to 11 V. When Q4 turns off again in state 4, the Q5g drops back down to 5 V. Q5
turns off again for state 1.
Appendix B. Additional waveforms for IC 75
Figure B.6: Switching nodes and low-side gating signals when operating in HSD modefor 42 V input voltage.
Appendix C
Efficiency Curves
Figure C.1: Additional efficiency curves using the inductors in the 0806 LR package
76
Appendix C. Efficiency Curves 77
Figure C.2: Additional efficiency curves using using the inductors in the 1210 size package
Appendix D
Alternate Level Shifter Topologies
One option for the level shifter is the simple pulsed SR latch depicted in Fig. D.1. When
the input goes high, a brief pulse is generated which turns the device Q1 high. This draws
current through the resistor R1, pulling the node labelled Vout low. This passes through
an inverter and sets the SR latch. For a reset the same thing happens with the second
branch. As this topology only draws current during a switching action, it can consume
a relatively low amount of power, as it consists of a small number of devices it will also
consume a relatively small amount of area. This topology is problematic however as it is
extremely susceptible to process, temperature, and voltage variations. Across a full set
of corners the swing can vary enormously. This can easily expose stages downstream of
the level shifter to overvoltage, or conversely create too small a swing to be registered
by the consecutive stage. The output swing for a variety of process corners is shown in
Fig. D.2. Nominally it is about 5 V, however, it can be as 7.5 V, and as low as 2.5 V,
depending on the value of VB and the process and temperature conditions.
78
Appendix D. Alternate Level Shifter Topologies 79
Figure D.1: Pulsed SR Latch based level-shifting topology
Figure D.2: Simulation showing output swing of pulsed SR latch across process andtemperature corners
This shifter can be modified to the one of Fig. D.3 where a PMOS whose gate is
biased to VS is placed in each branch, to prevent the output voltage from being pulled
below VS. Thus, instead of sizing the current and resistance value to generate the desired
swing, the pull-down devices are sized to achieve maximum voltage swing in the PVT
corner where the swing is smallest. The added PMOS devices prevent the voltage from
being pulled lower in corners where the swing is larger. A drawback of this is that it
does not have full voltage swing, as it can only pull the output as low as VS + Vthp,
Appendix D. Alternate Level Shifter Topologies 80
where Vthp is the threshold voltage of the PMOS device. This problem is exacerbated
by the fact that the threshold voltage also varies with PVT corners, leading to slightly
variable swing. This reduces the noise margin for the following stage as it must be able
to detect the smallest generated swing voltage as a low voltage. This topology also
has some other severe drawbacks. There is a significant tradeoff between the speed of
the shifter and its power consumption. As larger load resistors require less current to
generate the desired swing than smaller ones do, they consume less power. However, this
increases the RC time constant formed by the load resistor and parasitic capacitance at
the output nodes. If this time constant is too large, the circuit can becomes too slow to
respond to changes in VS and VB. This can cause a false triggering of the SR latch or
expose stages downstream of the level shifter to overvoltage as shown in Fig. D.4. In
addition to this, this design is relatively power hungry, consuming 3.5 mW in its worst
process corner. This is comparable to the gate drive losses of the stage that it is driving.
This topology is better suited when very high voltages are required as it can achieve
better ramp immunity (when combined with additional circuitry) then the level shifter
ultimately used, this was not necessary for the IC being developed, as the large relatively
large Coss values coupled with the modest switching node voltages limited the slew rate
of the switching node.
Figure D.3: Modified level shifter to limit swing to less than 5 V
Appendix D. Alternate Level Shifter Topologies 81
Figure D.4: Simulation showing level shifter perforamance across several corners
Also considered was the “bootstrapped” topology shown in Fig D.5. This topology
applies the concept of bootstrapping, usually used to power high-side gate drivers, to
the level-shifting itself. It consumes very little power. However it suffers from several
drawbacks. First, it may only be used if the source voltage is 0 at the moment of turn-on,
which makes it unsuitable for several of the switches in the 7SFC. Additionally, it will
not have full-swing, facing similar challenges to the protected latched topology. Hence it
was not an attractive design.
Appendix D. Alternate Level Shifter Topologies 82
Figure D.5: Schematic of bootstrapped level shifter
Bibliography
[1] Ipad pro 10.5 teardown. https://www.ifixit.com/Teardown/iPad+Pro+10.5-Inch+
Teardown/92534. Accessed : 2017-12-18.
[2] Andrija Stupar. Multi-Objective Optimization of Unidirectional Non-Isolated
DC/DC Converters. PhD thesis, University of Toronto, 10 King’s Colle Road, 7
2017.
[3] Henry Zhang. Basic concepts of linear regulator and switching mode power supplies.
Technical report, Linear Technology and Analog Devices, 2013. URL http://cds.
linear.com/docs/en/application-note/AN140fb.pdf.
[4] Gurjit Thandi Walt Kester, Brian Erisman. Switched capacitor volt-
age converters. Technical report, Analog Devices, 1998. URL
http://www.analog.com/media/en/training-seminars/design-handbooks/
Practical-Design-Techniques-Power-Thermal/Section4.pdf.
[5] Xunwei Zhou, Pit-Leong Wong, Peng Xu, Fred C Lee, and Alex Q Huang. Investi-
gation of candidate vrm topologies for future microprocessors. IEEE Transactions
on Power Electronics, 15(6):1172–1182, 2000.
[6] Brett Miwa. Interleaved conversion techniques for high density power supplies. PhD
thesis, Massachusetts Institute of Technology, 1992.
[7] David Baba. Benefits of a multiphase buck converter. Technical report, Texas
Instruments, 2012. URL http://www.ti.com/lit/an/slyt449/slyt449.pdf.
[8] Nenad Vukadinovic. Low-Power Multi-Level Flying Capacitor Converters - Mod-
elling and Control. PhD thesis, University of Toronto, 2017.
[9] Thierry A Meynard and Henri Foch. Multilevel converters and derived topologies
for high power conversion. In Industrial Electronics, Control, and Instrumentation,
83
BIBLIOGRAPHY 84
1995., Proceedings of the 1995 IEEE IECON 21st International Conference on, vol-
ume 1, pages 21–26. IEEE, 1995.
[10] K Nishijima, K Harada, T Nakano, T Nabeshima, and T Sato. Analysis of double
step-down two-phase buck converter for vrm. In Telecommunications Conference,
2005. INTELEC’05. Twenty-Seventh International, pages 497–502. IEEE, 2005.
[11] Parth Jain, Aleksandar Prodic, and Alexander Gerfer. Wide-input high power den-
sity flexible converter topology for dc-dc applications. In Applied Power Electronics
Conference and Exposition (APEC), 2016 IEEE, pages 2553–2560. IEEE, 2016.
[12] Zhenyu Zhao and Aleksandar Prodic. Continuous-time digital controller for high-
frequency dc-dc converters. IEEE Transactions on Power Electronics, 23(2):564–573,
2008.
[13] A Costabeber, L Corradini, P Mattavelli, and S Saggini. Time optimal, parameters-
insensitive digital controller for dc-dc buck converters. In Power Electronics Special-
ists Conference, 2008. PESC 2008. IEEE, pages 1243–1249. IEEE, 2008.
[14] Guang Feng, Eric Meyer, and Yan-Fei Liu. High performance digital control algo-
rithms for dc-dc converters based on the principle of capacitor charge balance. In
Proc. IEEE PESC Conf, pages 1740–1743, 2006.
[15] Guang Feng, Eric Meyer, and Yan-Fei Liu. A new digital control algorithm to achieve
optimal dynamic performance in dc-to-dc converters. IEEE Transactions on Power
Electronics, 22(4):1489–1498, 2007.
[16] Eric Meyer, Zhiliang Zhang, and Yan-Fei Liu. Digital charge balance controller to
improve the loading/unloading transient response of buck converters. IEEE Trans-
actions on Power Electronics, 27(3):1314–1326, 2012.
[17] Aleksandar Radic, Zdravko Lukic, Aleksandar Prodic, and Robert de Nie. Minimum
deviation digital controller ic for single and two phase dc-dc switch-mode power
supplies. In Applied Power Electronics Conference and Exposition (APEC), 2010
Twenty-Fifth Annual IEEE, pages 1–6. IEEE, 2010.
[18] Aleksandar Radic, Adrian Straka, and Aleksandar Prodic. Low-volume stackable
flyback converter with near minimum deviation controller. In Applied Power Elec-
tronics Conference and Exposition (APEC), 2014 Twenty-Ninth Annual IEEE, pages
1948–1953. IEEE, 2014.
BIBLIOGRAPHY 85
[19] Shadi Dashmiz, Behzad Mahdavikhah, Aleksandar Prodic, and Brent McDonald.
Hardware efficient auto-tuned linear-gain based minimum deviation digital controller
for indirect energy transfer converters. In Applied Power Electronics Conference and
Exposition (APEC), 2017 IEEE, pages 1994–1998. IEEE, 2017.
[20] Aleksandar Radic. Practical Volume-Reduction Strategies for Low-Power High-
Frequency Switch Mode Power Supplies. PhD thesis, University of Toronto, 2014.
[21] Arvind Raj. Calculating eficiency. Technical report, Texas Instruments., 2010. URL
http://www.ti.com/lit/an/slva390/slva390.pdf.
[22] Neil HE Weste and David Harris. CMOS VLSI design: a circuits and systems
perspective. Pearson Education India, 2015.
[23] Pcb layout techniques of buck converter. Technical report, Rohm, 2012.
URL http://rohmfs.rohm.com/en/products/databook/applinote/ic/power/
switching regulator/converter pcb layout appli-e.pdf.
[24] Yashodhan Moghe, Torsten Lehmann, and Tim Piessens. Nanosecond delay floating
high voltage level shifters in a 0.35 μm HV-CMOS technology. IEEE Journal of
Solid-State Circuits, 46(2):485–497, 2011.
[25] AN-1965 LM25011 Evaluation Boad. Texas Instruments, 10 2011.
[26] LMR24210/20 Evaluation Board. Texas Instruments, 10 2011.
[27] Walt Kester, Brian Erisman, and Gurjit Thandi. Voltage regu-
lator module (vrm) and enterprise voltage regulator-down (evrd)
11.0. Technical report, Intel Corp., 2009. URL https://www.
intel.com/content/dam/www/public/us/en/documents/design-guides/
voltage-regulator-module-enterprise-voltage-regulator-down-10-0-guidelines.pdf.
[28] Yuri Panov and Milan M Jovanovic. Design considerations for 12-v/1.5-v, 50-a
voltage regulator modules. IEEE Transactions on Power Electronics, 16(6):776–
783, 2001.
[29] Pit-Leong Wong, Fred C Lee, Xunwei Zhou, and Jiabin Chen. Vrm transient study
and output filter design for future processors. In Industrial Electronics Society, 1998.
IECON’98. Proceedings of the 24th Annual Conference of the IEEE, volume 1, pages
410–415. IEEE, 1998.
BIBLIOGRAPHY 86
[30] Kaiwei Yao. High-frequency and high-performance VRM design for the next gener-
ations of processors. PhD thesis, University of Toronto, 2004.
[31] Guang Chen, Ahmed Abou-Alfotouh, Zhiwei Liu, Mostafa Shabban, and Dan Oh.
Optimization of pcb pdn design using enhanced vrm model. In Electromagnetic Com-
patibility (EMC), 2014 IEEE International Symposium on, pages 845–849. IEEE,
2014.
[32] Aleksandar Radic, Adrian Straka, and Aleksandar Prodic. Synchronized zero-
crossing-based self-tuning capacitor time-constant estimator for low-power digitally
controlled dc–dc converters. IEEE Transactions on Power Electronics, 29(10):5106–
5110, 2014.
[33] Aleksandar Radic, David Baik, Adrian Straka, Aleksandar Prodic, and Robert
de Nie. Noninvasive self-tuning output capacitor time constant estimator for low
power digitally controlled dc-dc converters. In Applied Power Electronics Confer-
ence and Exposition (APEC), 2013 Twenty-Eighth Annual IEEE, pages 559–562.
IEEE, 2013.
[34] Adrian Philip Straka. Multi-mode near minimum deviation controller IC for high
frequency switch-mode power supplies. PhD thesis, University of Toronto (Canada),
2014.
[35] K Abe, K Nishijima, K Harada, T Nakano, T Nabeshima, and T Sato. A novel three-
phase buck converter with bootstrap driver circuit. In Power Electronics Specialists
Conference, 2007. PESC 2007. IEEE, pages 1864–1871. IEEE, 2007.