Prospects for Advances

i Power
in P Magnetics
M ti
C
Charles R. S
Sullivan
[email protected]

Dartmouth Magnetics and Power

Electronics Research Group

http://power.engineering.dartmouth.edu 1
Why Magnetics?
 Magnetics
g ((inductors and transformers)) are
required for most power conversion circuits, but
are responsible for much of the
 Size ((volume
Si l and
d weight)
i ht)
 Power loss
 Cost
 Difficulty in design (long development cycles)
 Magnetics are (almost) only used for power
 Less R&D than semiconductors.
 Less expertise among designers and users
users.

power.thayer.dartmouth.edu 2
Improvement avenues
 Better designs and design methods
methods.
 Better materials.
 Higher frequencies enabled by SiC and GaN.
GaN
(size 1/f ?)
 Or Si at low voltage; or circuit designs
designs…
 Materials and design needed for higher
frequencies.
 Higher frequencies enable new designs and
materials.

power.thayer.dartmouth.edu 3
Directions nott mutually
t ll exclusive
l i

Miniaturization:
V
Very hi h ffrequency,
high
integrated, thin-film, small,
and (eventually) low cost.

Ultra-high efficiency:
Efficiencies in the high
g
90’s without increasing
size or cost.

High-power
High-frequency,
high-efficiency at
m lti MW scale.
multi-MW scale

power.thayer.dartmouth.edu 4
Miniaturization vs
vs. efficiency
Reduced size at constant efficiencyy

Balanced benefits
Technology
Power allowing
Improved efficiency at constant size
density miniaturization

Th
Thermal
l limit
li it
size

loss Efficiency
 Because of thermal limits, miniaturization dividends
must be taken partly in efficiency.

power.thayer.dartmouth.edu 5
Miniaturization vs.
vs efficiency vs.
vs cost
Not an option

Short term: increased or constant cost

Long term: reduced cost
Technology
Power allowing
Increased cost
density miniaturization

Th
Thermal
l lilimitit
size

loss Efficiency
 Bottom line: any progress on cost, size, or efficiency can be used to
improve all of the above.
 In addition: Reduced cost facilitates adoption of efficient technology.

power.thayer.dartmouth.edu 6
Miniaturization theory and
practice
■ Simple theory: increase frequency
volume  mass  cost  1/f
■ Problem: power losses  fX
■ Windings: skin effect and proximity effect.
Skin Depth
Frequency 60 Hz 20 kHz 200 kHz 1 MHz 10 MHz
Skin depth, δ 8.5 mm 0.467 mm 0.148 μm 66 μm 21 μm
AWG for d = δ AWG 0 AWG 24 AWG 35 AWG 42 AWG 51

■ Cores: hysteresis, eddy-current, and “anomolous”

losses increase with frequency.

power.thayer.dartmouth.edu 7
 “Performance factor” for
MnZn power ferrites
ferroxcube
 Curves of BB·ff
product for 30 kHz·T
constant power
loss density.
density
 Looks like 200 mT

rapidly 20 kHz·T
accelerating
gains beyond 1
MHz!
 Shouldn’t
Sh ld ’t we
see frequencies
above 1 MHz plotted for 500 mW/cm3
rapidly
idl adopted?
d t d?

power.thayer.dartmouth.edu 100 kHz 1 MHz 8

Performance factor
factor, linear scale
90
80
70
T)

60
f x B ((MHz mT

50
40
30
20
10
0
0 1 2 3 4 5 6 7
f (MHz)
power.thayer.dartmouth.edu 9
Performance factor,
factor log scale
100
T)
f x B ((MHz mT

Slope = 1/2

10
0.01 0.1 1 10
f (MHz)
power.thayer.dartmouth.edu 10
Modified performance factor
 Accounts for worse winding loss at HF
HF.
 B x f 0.75 70
60
^0.75 mT MHz^0.75
5

50

40

30

20
f x B^

10

0
0.01 0.1 1 10
f (MHz)
power.thayer.dartmouth.edu 11
Material requirements for >1 MHz
 Performance factor must increase at least ~ f 0.5
 Goal: modified performance factor
> 100 (mT MHz 0.5 ) at 500 mW/cm3
 Consider higher power density spec for small components.
 Nothing commercial but much research activity.
 Some thin-film
S thi fil materials
t i l
 Many projects on ARPA-E ADEPT program
 IEEE Power System-On-Chip
y p Workshop p ((PowerSoC))
San Franciso Nov. 16-18.
 Many sessions on soft materials announced for
intermag/MMM Jan 2013
2013.

power.thayer.dartmouth.edu 12
Air core vs
Air-core vs. magnetic core

coil coil
Magnetic
Core:
soft
ft
magnetic
material
 “Air core”  Magnetic core with relative
is really any increases L (inductance) permeability
non-magnetic,
g by a factor μr, μr >> 1
dielectric typ. 4 ~10,000
material.  Allows smaller volume,
fewer turns for same L ((B = μr·μ
μ0·H))

power.thayer.dartmouth.edu 13
The case for air
air-core
core
 At veryy high
g frequencies
q
(e.g. VHF = 30 MHz to 300 MHz), the inductance value is
small and readily achievable without magnetic materials.
 Always gets better at higher frequency:
At constant η (constant Q): Volume proportional to f -3/2
At constant heat flux: Volume proportional to f -1/2
with Q improving as f 1/3
 Avoid magnetic material disadvantages:
 Power losses due to hysteresis and eddy currents.
 Frequency limitations: any material gets too lossy above some
frequency, often only a few MHz.
 Not available in standard IC or packaging processes.
 Can be used as a flotation device in case of flooding
flooding.

power.thayer.dartmouth.edu 14
Air core challenges
Air-core
 Low permeability leads to low flux density, which
requires substantial volume for sufficient flux.
 Requires more turns → more winding loss.
 Fl is
Flux i nott contained
t i d by
b a core; external
t l flflux can
cause EMI problems and eddy-current loss in nearby
conductors.
 Requires significant air volume.

power.thayer.dartmouth.edu 15
Air core geometries
Air-core
Our work
C.P. Yue,
CMU

Toroid Planar spiral

 No external flux  Flux in substrate,
problem. surroundings.
g
 Field is parallel to
 Field perpendicular to
conductor: low eddy
currents. conductor: high eddy currents.
 Easy to make.
 Harder to make.

 Q: can be > 100  Typical Q:

< 10 for 1 GHz

< 4 for 100 MHz
power.thayer.dartmouth.edu 16
 Air-core performance limits

100
um Q

 Field between two

layers of copper (e
(e.g.
g
Maximu

large-footprint toroid)

300 MHz 100 MHz 30 MHz

10

1
10 μm 100 μm 1
mm  Requires height for
Height
flux path!
power.thayer.dartmouth.edu 17
Air core: area requirement
Air-core:
140
 Data for: Solenoid
120 1:1 Solenoid
 3 mm height Toroid
 10 MHz y x
100
 For >10 MHz best design
1.3 X
and moderate 80
4X
miniaturization, 60
Q of b

air-core can
work. 40

 S l
Solenoidid may 20
require 4X
smaller area. 0
0.01
0 01 01
0.1 1 10 100 1000
2
area (cm )

power.thayer.dartmouth.edu 18
Options to improve vs
vs. air
air-core
core
 Multilayer copper
copper—each
each layer thin compared
to a skin depth. Can improve Q by a factor: p
where p is number of copper
pp layers.
y
 Magnetic material. Can improve by a factor μr,
the relative p
permeability,
y e.g.
g 10 to 10,000, if:
 Avoid saturation.
 Loss in the magnetic material is negligible.

power.thayer.dartmouth.edu 19
Thin film magnetic materials
Thin-film
 Typically alloys with Fe
Fe, Co and or NiNi.
 Sputtered or electroplated.
 R l ti permeability
Relative bilit μr in
i th
the 100’
100’s or 1000’
1000’s.
 Resistivity ~ 20 – 600 μΩ·cm;
 High end preferred for low eddy-current loss
 May still need laminations.
 Hysteresis loss?

power.thayer.dartmouth.edu 20
Magnetic anisotropy
 Hard axis loop provides:
 Low
1 5 permeability
1.5
needed to avoid Easy Axis
saturation
1.0 in inductors.
 Low
L h
hysteresis
t i lloss.
0.5
Hard Axis:
M ((T)

00
0.0 Near perfect
Near-perfect
lossless loop
-0.5

-1.0

-1.5
-1 5
-200 -100 0 100 200
Field (Oe)
power.thayer.dartmouth.edu 21
Nano composite magnetic materials
Nano-composite
Magnetic Metal
(3~5 nm Co
Particles)

Ceramic (Al2O3, ZrO2, etc.)

Advantages:
 Ferromagnetic (coupled particles)

 High resistivity (300 ~ 600 μΩ·cm) controls eddy-

currentt loss
l independent
i d d t off flflux di
direction.
ti
 Some have strong anisotropy for low permeability

and low hysteresis loss.

power.thayer.dartmouth.edu 22
Permeabilty vs.
vs frequency
1000


100  c
Q = 370 Q = 37
eability

10  
Perme

 c
1

01
0.1
1 10 100 1000 10000
Frequency ( MHz )
power.thayer.dartmouth.edu 23
TEM of multilayer nanogranular film
Co
C
O
Si

HRTEM11

100 nm

power.thayer.dartmouth.edu 24
Modified Performance Factor
 Preliminary results on a single sample
sample….
300
Ni-Fe-Zr-O
^0.75 mT MHz^0.75
5

250

200

150

100
f x B^

Ferrites
50

0
0.01 0.1 1 10
f (MHz)
power.thayer.dartmouth.edu 25
T
Two types
t off inductors
i d t
Pot-core Toroidal

 Core wraps winding  Winding wraps core

power.thayer.dartmouth.edu 26
 Microfabricated inductors
Pot-core Toroidal
(MCM) (CMC)

 Two magnetic depositions  Does not work with

 Uses magnetic material uniaxial anisotropy
only in hard axis
power.thayer.dartmouth.edu 27
Racetrack inductors fabricated at Dartmouth
18

16

14
12

Quality factor
10

6 7-turn inductor A
7 turn inductor B
7-turn
4 8-turn inductor A
2 8-turn inductor B
Previous 7-turn
0
1 10 100
Frequency (MHz)
1000

800

Inductance (nH)
600

400
7-turn inductor A
7-turn inductor B
200 8-turn inductor A
8-turn inductor B
Previous 7-turn
0
2/25/2013 1 10 100 28
Frequency (MHz)
Other inductors
we’ve
we ve developed
 V-groove
V groove 1
1-turn
turn inductor
for high current (up to 12 A)
 Polyimide
P l i id substrate
b t t withith sputtered
tt d
material on both sides
 Microfabricated coupled inductors
(2004, with Tyndall)

power.thayer.dartmouth.edu 29
 Performance:
power density and efficiency
efficiency.
 2011 V-groove:
V groove: 2011
V-groove
measured
inductor with 2005 2012
calculated
l l t d circuit
i it V-groove Racetrack

 2012 Racetrack:
100 V in,
36 V out,
First test—
process in
progress.

power.thayer.dartmouth.edu 30
Flux crossing magnetic laminations

 Problem in corners
where top and bottom
magnetic core halves
j i
join.
 Excess eddy currents
limit efficiency and Q.

31
Solution: Toroidal inductors
 Advantage:
 Flux stays in plane, minimizing eddy-current losses.
 Challenge:
 Flux direction varies;; sometimes oriented incorrectlyy for the
magnetic material anisotropy.
 Solution:
 Induced radial anisotropy,py, such that flux travel is always
y in the
low-loss hard-axis direction.
Winding Radial anisotropy

Core
Flux

32
Concept

 Use permanent-magnet
permanent magnet array to apply field
during deposition
ring
i magnets
t di magnets
disc
iron sheet

33
 Fixture to deposit toroidal cores with
radial anisotropy
1st g
gen. Toroidal core size:
substrate
19 mm 
copper

iron
8 mm
ID=8 mm, OD=16 mm

magnet 12 4 mm
12.4

8 mm
25 4 mm
25.4

Polyimide
substrate

Copper

Iron

Magnets

power.thayer.dartmouth.edu
Iron 34
 Hysteresis measurements confirm
radial anisotropy

A square sample about 1 mm

Square by
sample 1 mm was cut for
B-H loops
p with a radial field: p without a radial field:
B-H loops
measurements
1.5 1.5
easy-axis loop
radial loop
hard-axis loop
1 hard-axis minor loop 1 1 circumferential loop
hard-axis minor loop 2

0.5 0.5

B (unncalibrated)
ncalibrated)

0 0
B (un

-0.5 -0.5

-1 -1

-1.5 -1.5
-200 -150 -100 -50 0 50 PowerChip
100 -200
150 - Integrated
200 -150 -100
Power -50 0 50 100 150 200
drive field (Oe) Electronics drive field (Oe) 35
Complex permeability of toroidal
samples

1000

relatiive permeaability

Q r

100
 
r
10

0.1
10 100 1000 3000
f (MHz)

36
Ultra high efficiency
Ultra-high
Miniaturization:
V
Very hi h ffrequency,
high
integrated, thin-film,
small, and (eventually)
low cost.
cost

Ultra-high efficiency:
Efficiencies in the high
g
90’s without increasing
size or cost.

High-power
High-frequency, high-
efficiency at multi-MW
scale
scale.

power.thayer.dartmouth.edu 37
Ultra high efficiency
Ultra-high
 Accurate loss models are essential:
 Understand all sources of loss.
 Optimize to get ultra-high
ultra high efficiency without high
cost.
 Co-design with circuits.
 Frequency and material choice:
 Magnetic material for > 1 MHz?
 Probably < 1 MHz, using advanced
semiconductors for low loss not high frequency.

power.thayer.dartmouth.edu 38
Loss modeling and optimization
 Cores: data sheets report loss with sine
waves, but power converters use square
waves. Nonlinear behavior means Fourier
analysis doesn’t help.
 Methods available but more research needed.
 Windings:
Need wire much smaller than a skin depth.

power.thayer.dartmouth.edu 39
Present best technology for low
winding loss: litz wire

Image: Noah Technologies

power.thayer.dartmouth.edu 40
Litz wire progress
 History
 Invention: 1888, Sebastian de Ferranti.
 A l i 1917 H
Analysis Howe; 1926 BButterworth.
tt th
 Optimization, 1997, 1998; CAD tools 2003.
 Low loss up to ~1 MHz.
 Needs:
 Extend to 10~100 MHz.
 Lower cost.

power.thayer.dartmouth.edu 41
 Litz wire design options
Litz-wire
10
32

34

36
L
Loss
rmalized Loss

38
Full-bobbin design—
40 double the loss of cheaper
Norm

design.
42
1
Cheaper and 44
Good design
lower loss 46
requires optimization
48
50
0.1 1 10 100
Cost….~ amount of Cu
Normalized Cost

power.thayer.dartmouth.edu
Litz > 1 MHz
 Available improvement vs
vs. single
single-layer
layer
solid wire:

60 Hz 20 kHz 200 kHz 1 MHz 10 MHz

8.5 mm 0.467 mm 0.148 μm 66 μm 21 μm
AWG 0 AWG 24 AWG 35 AWG 42 AWG 51

 Factor of 3
3.6
6 with AWG 48 at 1 MHz
 Factor of 1.4 at 10 MHz with AWG 50

power.thayer.dartmouth.edu 43
Litz > 3 MHz
 Goal: dimensions smaller than 15 μm
 Possibilities:
 Improved wire drawing and handling.
 Foils and films…might need to interchange
layers.
 Research goal: achieve current sharing
b t
between layers
l without
ith t iinterchanging.
t h i

power.thayer.dartmouth.edu 44
Other metals?
Metal Resistivity ρ (conductor
( d t grade)
d )

Silver (Ag) 1.59 Ω·cm

C
Copper (C
(Cu)) 1 72 Ω·cm
1.72 Ω
Gold (Au) 2.44 Ω·cm
Aluminum (Al) 2.83 Ω·cm
 Note units: cm not kg g
 Price per kg is irrelevant:
consider price per volume
volume.

power.thayer.dartmouth.edu
Costs: Cu vs
vs. Al
 Mass:
$9.20/kg vs. $2.40/kg (Irrelevant)
 Volume:
8.2 ¢/cm3 vs. 0.65 ¢/cm3 12X
 Resistance:
$Ω $Ω
14 m2 vs. 1.8 m2 7.7X
 Actual Al magnet
g wire p
prices aren’t that g
good,, but
even at the same price per kg, Al is ½ the price for
the same resistance.

power.thayer.dartmouth.edu
Can Al work at high frequency?

power.thayer.dartmouth.edu 47
 Measured transformer ac resistance

Rac

Al 28%
worse
(30%
measured)
 29 turns of 0.5
mm (AWG 24) Al
wire. Al no worse
 T
Two layers
l on
EE19 core. Cu
Al 64% worse (61% measured)

power.thayer.dartmouth.edu
Cost and loss of Al vs.
vs Cu
 For same dc resistance
resistance, Al is >7X cheaper
(commodity price).
 Advantages at high frequency are even
greater than at low frequency.
 Only reasons for Cu:
 Where compact size is more important than
efficiency,
y, cost,, temperature
p or weight.
g
 If termination cost difference exceeds wire
cost difference.

power.thayer.dartmouth.edu
Directions Miniaturization:
atu at o
• Higher frequency
• Thin film
microfabrication
• Ferrites > 10 MHz,
conductors > 10 MHz

Ultra-high
Ult hi h efficiency:
ffi i
Materials and
techniques to lower
cost and optimize
designs

High-power
High-frequency, high-
efficiency at multi-MW
scale
scale.

power.thayer.dartmouth.edu 50
High Power
Miniaturization:
V
Very high
hi h ffrequency,
integrated, thin-film,
small, and (eventually)
low cost.
cost

Ultra-high efficiency:
Efficiencies in the high
g
90’s without increasing
size or cost.

High-power
High-frequency, high-
efficiency at multi-MW
scale
scale.

power.thayer.dartmouth.edu 51
High power characteristics
High-power
 Large size makes two particular
challenges
h ll severe:
 Overall winding size is large compared to
skin
ki ddepth.
th 10 cm scalel att 10 kH
kHz h
has th
the
same ratio (150) as mm scale for 100 MHz.
 Th
Thermal l management—low
t l surface-area
f
to volume ratio.
 Nanocrystalline
N t lli magnetic ti materials:
t i l
better than ferrite at < 100 kHz, high flux
density.
densit
power.thayer.dartmouth.edu 52
Existing magnetics
10 MW

1 MW

100 kW

10 kW

1 kW

100 W

10 W
Emerging
1W thin-film
60 Hz 1kHz 10 kHz 100 kHz 1MHz 10 MHz 100 MHz

power.thayer.dartmouth.edu 53
Needed with new semiconductors
10 MW

1 MW

100 kW

10 kW

1 kW

100 W

10 W
Emerging
1W thin-film
60 Hz 1kHz 10 kHz 100 kHz 1MHz 10 MHz 100 MHz

power.thayer.dartmouth.edu 54
Projected capability
10 MW

1 MW

100 kW

10 kW

1 kW

100 W

10 W
Emerging
1W thin-film
60 Hz 1kHz 10 kHz 100 kHz 1MHz 10 MHz 100 MHz

power.thayer.dartmouth.edu 55
Research opportunities
10 MW

1 MW
New magnetic
100 kW materials and
conductor
10 kW apporaches?

1 kW ?
100 W

10 W
Emerging
1W thin-film
60 Hz 1kHz 10 kHz 100 kHz 1MHz 10 MHz 100 MHz

power.thayer.dartmouth.edu 56
Selected references
Tutorial/overview on integrating magnetics for on-chip power converters with 66 references and
more extensive discussion of air-core and magnetic-core miniaturized power magnetics:
 C.R. Sullivan, “Integrating Magnetics for On-Chip Power: Challenges and Opportunities,”
Invited paper. IEEE Custom Integrated Circuits Conference, Sept. 2009.
Ad
Advanced d modeling
d li off losses
l in
i multilayer
ltil thi
thin-film
fil materials,
t i l andd also
l iintroduces
t d a modified
difi d
performance factor for magnetic materials that is more appropriate for high-frequency designs:
 Di Yao and C.R. Sullivan, “Effect of Capacitance on Eddy-Current Loss in Multi-Layer
Magnetic Films for MHz Magnetic Components,” IEEE Energy Conversion Conference and
Exposition Sept.
Exposition, Sept 2009.
2009
Measurements of losses in ferrites with non-sinusoidal waveforms that show phenomena not
expected based on existing models:
 C.R. Sullivan, J.H. Harris, E. Herbert “Core Loss Predictions for General PWM Waveforms
from a Simplified
p Set of Measured Data,”, IEEE Applied
pp Power Electronics Conference,, Feb.
25, 2010.
A high-level theoretical overview of the potential for reducing winding loss:
 M.E. Dale and C.R. Sullivan. “Comparison of Single-Layer and Multi-Layer Windings with
Physical Constraints or Strong Harmonics.” IEEE International Symposium on Industrial
Electronics, July 2006.
Aluminum windings for high-frequency applications:
 C.R. Sullivan, "Aluminum Windings and Other Strategies for High-Frequency Magnetics
Design in an Era of High Copper and Energy Costs." IEEE Transactions on Power
El t i
Electronics, 23(4) JJuly
l 2008

power.thayer.dartmouth.edu 57