1

Fundamental Examination of Multiple Potential

Passive Component Technologies for Future Power
Electronics
Phyo Aung Kyaw, Student Member, IEEE, Aaron L.F. Stein, and Charles R. Sullivan, Fellow, IEEE

Abstract—Efficient and compact high-power high-frequency storing electrical energy on the time scale of the switching
passive energy storage components are required for miniatur- frequency, other energy storage mechanisms may be used as
ization of power converters, and remain a challenging obstacle alternatives to the magnetic and electrostatic passives in power
in power electronics. In this work, multiple energy storage
mechanisms are analyzed on an order-of-magnitude basis to converters. Noworolski and Sanders [5] proposed replacing the
identify potential alternatives for conventional passive compo- conventional LC tank with an electrostatically coupled micro-
nents, especially magnetics, which have frequencyyeah-dependent fabricated microresonant structure. Piezoelectric transformers,
power losses. The high energy density of mechanical storage vibrating piezoelectric bars tapped at various positions for
methods provides an attractive alternative to the widely-used LC voltage transformation [6], can be designed in various types
resonance. Piezoelectric transduction is explored and the perfor-
mance of a piezoelectric resonator is compared to that of an LC and structures. They are used in place of electromagnetic trans-
resonator in terms of efficiency and power handling capabilities in formers for applications such as fluorescent lamp inverters [7],
a resonant switched-capacitor type circuit. The analysis provides [8], ac-adapters [9] and power supplies [10] for power ranges
a basis for exploring potential passive energy storage component on the order of 10 W.
technologies and comparing their performance limits with those
of electromagnetic passive components. The analysis shows that In this paper, we present a first-order analysis of energy
both electromagnetic and mechanical resonance, in the ideal densities of various mechanical and electrical energy storage
scenarios, can offer much better performance than do passive mechanisms to identify potential passive component technolo-
component technologies in use today. A prototype resonator is
built with off-the-shelf C0G capacitors to confirm the low loss gies for power conversion at MHz frequencies. Kinetic, elastic,
and high power predicted by the model. The resulting resonator electrostatic and magnetic energy storage are considered. To
has an effective resistance of 2.76 mΩ at 2.70 MHz in 1 cm3 analyze the upper performance limit of each mechanism,
volume and can handle 7.42 kW within the rated temperature. we assume best-case scenarios, considering mainly the lim-
Index Terms—energy storage, piezoelectricity, resonators, res- its imposed by the properties of high-performance materials
onant power conversion, switched capacitor circuits currently available for each method, and ignoring the practical
limitations such as those regarding current fabrication technol-
I. I NTRODUCTION ogy. The goal of this work is to identify opportunities for tech-
Switching power converters require passive components, nology development for future power conversion components
such as inductors, transformers and capacitors, which often and systems and not to provide a design process for power
dominate the size and loss of power converters. Recent ad- converters. Realizing the potential identified by such analysis
vances in wide-bandgap semiconductors such as GaN and SiC may require extensive additional research and development
allow for efficient high-frequency power conversion [1], and effort.
further improvement to power semiconductor technology may The analysis of energy density shows that mechanical
be expected. The major limitations impeding further improve- energy is attractive, but it requires transduction for use in
ment of power converter density are the densities and power electrical power converters. To consider this, we analyze
losses of passive components, especially magnetics [2], [3]. piezoelectric transduction to couple mechanical and electrical
Although incremental improvements in magnetics are enabling energy. The Butterworth-Van Dyke circuit model for a piezo-
advances in power density, the importance of the problem electric resonator is described and used in the analysis. To
merits consideration of the fundamental performance limits compare the performance of piezoelectric resonators to that of
of passive components and exploration of possible alternative conventional LC resonators, a simple 1:1 resonant switched-
technologies. capacitor (ReSC) type circuit is used as an example power
Recent research activities focus on improving magnetics application to standardize the analysis. This work extends
technology for the MHz frequency range, including optimiza- our conference publication [11] by using a more accurate
tion of inductor geometries and nanoengineering of mag- inductor model, considering skin effect in inductor windings
netic materials [4]. Because passive components are used for and commercially available low-energy-density dielectric, and
adding a more extensive study of power output capability
This work was supported by the National Science Foundation under Grant
Number 1507773 and 1610719. in a sample application. The results for the LC resonators
The authors are with the Thayer School of Engineering, Dartmouth with commercial C0G capacitors are then verified using an
College, Hanover, NH 03755 USA (e-mail: [email protected]; experimental prototype; the details of the design choices and
[email protected]; [email protected])
Color versions of one or more of the figures in this paper are available the small-signal experimental verification are presented in
online at http://ieeexplore.ieee.org. another conference publication [12].
 2

TABLE I
H IGH E NERGY D ENSITY C APACITOR D IELECTRIC M ATERIALS

Dielectric Material r E(V/µm) u(J/cm3 ) Ref

polypropylene 2.2 820 6.5 [14]
alkali-free glass a 6 1200 38 [14]
P(VDF-CTFE)/BNNS/BT b 12 552 21.2 [15]
PVDF/ArPTU c 9.2 700 10.8 [16]

a barium boroaluminosilicate
b poly(vinylidene fluoride-co-chlorotrifluoroethylene)/hexagonal boron ni-
tride nanosheets/barium titanate
c polyvinylidene fluoride/aromatic polythiourea

For elastic potential energy storage, the energy density

depends on the material strength as well as the stiffness. The
maximum elastic energy density of a rod is σy2 /(2Y ) where
Fig. 1. Comparison of achievable energy densities for different energy storage Y is the Young’s modulus of the material. For helical springs,
mechanisms using high performance materials. The energy densities are
independent of frequency, with the exception of inductors with a single-layer torsional yield strength and shear modulus need to be used in
winding whose energy density depends on the skin depth. The energy densities place of σy and Y respectively. Geometric factors such as the
are limited by material properties or power rating. The normal supercapacitors spring index (the ratio of the diameter of the spring coil to that
are limited by the response frequency; the higher the response frequency, the
lower the achievable energy density. of the wire) and the ratio of the volume of the spring material
to that of the box containing the spring also affect the energy
II. A NALYSIS OF ENERGY DENSITY density [11]. For steel, with σy of 150 to 1500 MPa and Y of
around 200 GPa [13], the achievable energy densities are 56
Kinetic, elastic, electrical and magnetic energy storage to 5600 mJ/cm3 for rods and 6.3 to 630 mJ/cm3 for helical
mechanisms are analyzed in terms of the maximum possible springs as shown in Fig. 1.
volumetric energy density u. The energy density is based on
the volume of the enclosing box of the storage element rather
B. Electrostatic Energy
than the volume of the material. For the preliminary analysis in
this section, ideal scenarios are considered in order to find up- Capacitors’ maximum electrical energy density is given by
per limits to possible performance that could be achieved with E 2 /2, where  is the permittivity and E the breakdown elec-
further technological development. These idealizations include tric field of the dielectric. Polypropylene, a common low-loss
mechanical oscillation in vacuum, zero capacitor dielectric loss polymer dielectric, has a maximum energy density of around
and zero magnetic core loss, but conduction loss is considered 6.5 J/cm3 with r = 2.2 and E = 820 V/µm [14]. Recent ef-
for magnetic energy storage. Later sections explore practical forts in enhancing the capacitor energy density involve forming
considerations that reduce performance relative to these ideal nanocomposites of high-breakdown-field polymers and high-
assumptions. In order to perform a fair comparison, a fixed dielectric-constant ceramics [15], [16]. Testing commercially
1 cm3 volume, 300 V rating and 10 MHz frequency are used available glass [14] also shows that alkali-free glass has a
to standardize the analysis. This section will summarize the high energy density. Table I compares the performance of
energy density analysis; details are discussed in [11]. polypropylene and state-of-the-art dielectrics in the literature;
these are the energy densities at the breakdown limits and
derating is necessary for continuous long-term operation.
A. Mechanical Energy Electrolytic double-layer supercapacitors are another way to
Kinetic energy stored in mechanisms such as a linearly achieve a high energy density. A permeable separator soaked
oscillating mass or a flywheel is proportional to the square of with an electrolyte is placed between electrodes with a large
the speed of oscillation or rotation. However, a faster motion surface area to volume ratio. The opposite charges are sepa-
requires larger restoring or centripetal forces. This presents rated by an ion-level thickness between the electrode surface
an upper limit to the achievable speed, and hence the energy and the electrolyte. Graphene is widely used in supercapacitors
density, depending on the compressive or tensile yield strength due to its large specific surface area of 2675 m2 /g [17]. This
of the material σy . The energy densities are given by σy /2 for large surface area means that supercapacitors have very high
a flywheel and σy /6 for an oscillating mass in vacuum. The energy densities, as much as 1 kJ/cm3 ; however, they usually
factor of three difference is due to the extra space required have large RC time constants on the order of milliseconds to
for oscillation; the difference will be larger if the air drag seconds [18], which is too slow for MHz frequency switching.
is considered [11]. For steel, one of the strongest materials, For frequencies higher than 1 MHz, the maximum achievable
with strength 150 to 1500 MPa [13], the maximum achievable energy density is about 20 J/cm3 .
energy density is 75 to 750 J/cm3 for flywheels and 25 to
250 J/cm3 for oscillating masses in vacuum. These energy C. Magnetic Energy
density ranges are summarized in Fig. 1 together with those Inductors’ maximum energy density depends on the per-
of other mechanisms discussed below. meability µ and the saturation flux density Bsat of magnetic
 3

cores, usually below 1 T. Much higher magnetic flux densities

are achievable with air-core inductors but at the expense of
high current and conductor losses. Pulsed power systems can
have an energy density as much as 39 J/cm3 with a maximum
flux density of 50 to 100 T, with the only limitation being
“the onset of melting at the conductor surface” [19]. However,
such a constraint does not apply in our study of storage
mechanisms for use in switching power converters at MHz Fig. 2. Sketch of an inductor with a foil winding.
frequencies since the high temperature will result in failure
of the insulation materials. Moreover, the temperature rise of TABLE II
a pulsed power system is limited by thermal mass and the M AGNETIC E NERGY S TORAGE D ENSITIES
same limit could lead to much higher temperatures in the
continuous operation of switching converters. The power loss Single-Layer Winding Multi-Layer Winding
Material
at a current density high enough to bring about melting of µopt u (mJ/cm3 ) µopt u (mJ/cm3 )
the conductor surface will also be too high to be useful in an Air 1 0.015 1 1.87
efficient switching converter. Bsat = 0.32 T 42 0.97 4 5.8
Bsat = 1 T 11 3 12 17.2
Thus, power loss in the continuous operation mode, and the
resulting temperature rise, is an important limit to consider.
In this paper, we use tolerable power loss per unit area by using an effective permeability µopt = Bsat /µ0 Hmax . Par-
Pmax = 3 W/cm2 , estimated based on a survey of com- ticular values for µopt can be physically attained by choosing
mercially available millimeter-scale surface mount resistors in the correct air gap length. For air-core inductors, the saturation
the Digi-Key online catalog [20]. The catalog was scanned flux density Bsat is simply µ0 Hmax .
to search for resistor series with a clear correlation between At MHz frequencies, the skin depth δ is on the order of tens
power rating and surface area; these correlations range from of micrometers, which limits the effective conductor thick-
1/2 3/2
Prating ∝ Asurf ace to Prating ∝ Asurf ace . For a majority of ness. However, this skin effect can be partially overcome by
these series, Prating ∝ Asurf ace , ranging from 1 W/cm2 to using multilayer foil windings [21]. Assuming optimistically
4 W/cm2 . We choose 3 W/cm2 as the tolerable power loss to that a multilayer technique works perfectly without requiring
use in the analysis. This is an aggressive assumption, based on additional space or incurring any excess power losses, the
resistors which typically tolerate high operating temperatures. result could conceivably be equivalent to allowing full use
However, as discussed below, inductors have a low energy of a conductor much thicker than a skin depth. In this
density even with this assumption of a high tolerable loss. case, the conductor thickness can be optimized for maximum
The power loss in an inductor winding (Fig. 2) is given by energy density subject to the power loss density constraint.
2wL For both single-layer and multi-layer cases, we calculate, and
P = I 2 ρe , (1) tabulate in Table II, the energy densities for some hypothetical
lL t L
lossless magnetic cores, representing ferrite (Bsat = 0.32 T),
where I is the current through the winding, ρe the resistivity, and metal and metal-alloy based materials (Bsat = 1 T) [22].
wL the inductor width, lL the inductor length and tL the Fig. 1 shows a comparison of achievable energy density
conductor thickness that is actively conducting current, which ranges for the storage mechanisms discussed above, limited
may be limited by skin depth as discussed below. For this anal- by the properties of high performance materials. Flywheels
ysis, we assume that the design is configured for negligible- and oscillating masses provide the highest energy density,
proximity-effect loss, and that the core loss is negligible, in about an order of magnitude higher than capacitors, which
order to determine an upper limit on possible performance that in turn have an order of magnitude better performance than
might be achievable with improved magnetic materials. Using springs. Inductors provide the lowest energy density of all
(1), the maximum magnetic field Hmax for a tolerable power the mechanisms, about two orders of magnitude lower than
loss per unit area Pmax can be obtained: springs. It should be noted that the energy density of inductors
P I 2 ρe is calculated using power rating of commercially available re-
Pmax = = 2 sistors in continuous operation whereas other mechanisms are
2 lL wL lL t L
s analyzed using electrical or mechanical breakdown. However,
Pmax tL even with an order of magnitude derating on other mechanisms
Hmax = . (2)
ρe for continuous operation, inductors’ energy density will still
For a hypothetical lossless magnetic material with a saturation be about an order of magnitude lower than that of springs.
flux density Bsat , the winding-loss-limited energy density is
  III. R ESONANCE , T RANSDUCTION AND E XAMPLE P OWER
1 tL 1 A PPLICATION
u= dL + Bsat Hmax , (3)
dL + 2tL 3µopt 2
Typical switching converters for MHz frequencies use an LC
where the fraction tL /(3µopt ) is to account for the energy resonant tank. An alternative resonant structure with higher
stored in the conductor. This energy density can be achieved energy density and lower losses could enable smaller, more
 4

efficient converters. In an LC or a mass-spring (MK) resonator,

the element with the lower energy density typically limits
the performance: the inductor in the LC resonator and the
spring in the MK resonator. Given that springs have an energy
density limit about 300 times higher than that of inductors, MK
resonance systems can be better than LC tanks at compactly
storing energy with MHz resonance frequencies.
Electromechanical transduction methods need to be inves-
tigated to exploit this high-energy-density MK resonance for
use in switching power converters. Among electromagnetic,
electrostatic and piezoelectric transduction systems, the first
can be ruled out since the necessary windings are subject to the
same performance limits as inductor windings. Electrostatic
Fig. 3. (a) A piezoelectric resonator with vibration and poling along the
systems, though similar to capacitors, have low energy density same direction (along d). (b) Butterworth-Van Dyke model of a piezoelectric
because of the lack of dielectric. Piezoelectric systems are resonator. C0 is the capacitance between the two electrode plates and R, L
interesting since they can drive oscillating masses and have and C are motional impedances.
a similar energy density limit to that of springs. Although
flywheels have the highest energy density of all the mecha- TABLE III
nisms discussed (Fig. 1), to our knowledge, there is not yet an L ITHIUM N IOBATE M ATERIAL P ROPERTY C ONSTANTS
efficient rotary piezoelectric driving mechanism for the MHz T Relative permittivity 28
r
frequency range [23]–[25]. Thus, we study the performance of
Y Young’s Modulus 2 × 1011 N/m2
a piezoelectric resonator to use in an MK resonance system
ρ Density 4700 kg/m3
with an oscillating mass. A resonant switched-capacitor type
ν Poisson’s Ratio 0.241
circuit [26], [27] is used as an example power application to
f0 Q Figure of Merit (Maximum f0 Q) 1 × 1014
compare the performance of the piezoelectric resonator to that
E Breakdown electric field 2.6 × 107 V/m
of a conventional LC resonator.
Ty Yield stress 1.1 × 108 N/m2
A. Piezoelctric Resonator Sy Yield strain 5.5 × 10−4
Fig. 3 (a) shows a piezoelectric resonator in length-
extensional vibration mode, in which the direction of vibration decrease in the length l0 increases the resonance frequency and
and poling are the same, and it can be described by the decreases the quality factor. Combining (8) and (9) results in
Butterworth-Van Dyke model shown in Fig. 3 (b). For a
resonator vibrating and poled in the thickness d direction, the Y
f0 Q = . (10)
capacitance C0 and the motional impedances R, L and C are 2πηρm
A This product is a function of the material properties and
C0 = , (4) is independent of the resonator dimensions, so it can be
d
π2 d π 2 ηρm considered as the figure of merit (FoM) for a piezoelectric
R= 2
ηρm = 2 , (5) material. Some research in piezoelectric materials focuses on
8k Y A 8k C0 Y
achieving a high f0 Q FoM, and values as high as 1.1 × 1014
d 1 ρm l02
L= 2 ρm l02 = 2 , (6) have been achieved by some groups [29], [30].
8k Y A 8k C0 Y
Lithium niobate is chosen as the material for analysis due to
8k 2 Y A 1 8k 2 C0
C= = , (7) its high electromechanical coupling coefficient and FoM [29],
2
π d Y π2 [31]; its properties are summarized in Table III [29], [30],
where A is the poling area (bl in Fig. 3 (a)), η the kinematic [32]–[35]. For a 1 cm × 1 cm × 1 cm block of lithium niobate,
viscosity of the material and the environment, ρm the density vibrating and poled in the same direction (parallel mode), with
of the material, Y the Young’s Modulus, k 2 the electrome- an optimistic assumption of k 2 = 0.3, the Butterworth-Van
chanical coupling coefficient and l0 the length of vibration (d Dyke model gives R = 2.6 mΩ, L = 0.40 H, C = 0.60 pF
for the longitudinal vibration in Fig. 3 (a)) [28]. and C0 = 2.5 pF. The resonator has a series resonance at
The resonance frequency f0 and the quality factor Q derived 326 kHz and a parallel resonance at 461 kHz.
from (4) – (7) are For vibration and poling in distinct directions (normal
s mode), lower impedance can be achieved with the same
1 1 Y
f0 = √ = 0 , (8) resonance frequency and volume. For operation in the normal
2π LC 2l ρm mode, the resonator in Fig. 3 (a) can be poled along the
s
same direction d with vibration along either l or b. The same
r
1 L l0 Y
Q= = . (9) resonance frequency as that for the parallel mode can be
R C πη ρm
obtained by adjusting the vibration length to account for the
There is a trade-off between the resonance frequency and difference in the material property constants between the two
the quality factor with respect to the length of vibration. A modes. The impedance can be lowered by reducing the poling
 5

We have explored various electromechanical transduction

methods to exploit the high energy density offered by MK
resonance and ruled out the electromagnetic and electrostatic
transduction mechanisms respectively due to the winding loss
and the low energy density. We also studied the piezoelec-
tric resonator and described an equivalent circuit model. In
addition, a ReSC circuit is described as an example power
application which can be used to compare the performance of
Fig. 4. 1:1 resonant switched-capacitor type circuit used for comparing the the piezoelectric and the conventional LC resonator.
mechanical and the electromagnetic resonances.
The two resonators are compared in the following sections.
In the analysis, an air-core inductor and a capacitor with alkali-
thickness to the minimum required to sustain the required free glass dielectric [14] are used for the LC resonator and
voltage rating, and connecting multiple layers in parallel. The lithium niobate [32] is used for the piezoelectric resonator. A
smaller poling thickness d increases the capacitance C0 and dc voltage rating of 300 V is assumed for both resonators.
reduces the motional resistance R while parallel connections
allow for further impedance reduction. Some of the practical
IV. R ESONANT S WITCHED -C APACITOR T YPE C IRCUIT:
limitations of this approach are discussed in Section IV-A.
E FFECTIVE R ESISTANCE
B. Resonant Switched-Capacitor Type Circuit A. Effective Resistance of a Piezoelectric Resonator
To determine the potential of the piezoelectric resonator, its The effective resistance of a piezoelectric resonator
performance needs to be compared to that of a conventional (Fig. 3 (a)) in a ReSC-type circuit at a resonance frequency
LC resonant tank. The performance metrics for comparison f0 can be derived from (11)–(13):
are losses in the resonator, and the power output limited by   −1
π
these losses or other mechanical nonlinearities. As a standard Ref f,piezo = f0 C0 + f0 C coth . (15)
4Q
for comparison, we use a 1:1 resonant switched-capacitor
(ReSC) circuit (Fig. 4) with 300 V nominal input and output Assuming a high quality factor (Q  1) and using (4) and
voltages [26], [27]. Although we use a 1:1 converter for (7), the expression can be simplified:
simplicity, the trends found would apply to a variety of ReSC 1

8k 2 4Q
−1
applications. ReSC converters have recently attracted interest Ref f,piezo ≈ 1+ 2
f0 C0 π π
and have shown to be surprisingly versatile [36]–[40]. For
example, [39], [40] demonstrate that they can be operated in π2 1 π2 π2 R
≈ 2 = , (16)
modes that provide continuous variation of the output voltage. 8k C0 2πf0 Q 2 2
The process of charge transfer between the input VA and where R is the motional resistance described in (5). Thus, the
the output VB can be characterized by the effective resistance effective resistance of a high Q piezoelectric resonator in a
Ref f in Fig. 4. This effective resistance is given by ReSC-type circuit depends only on the capacitance C0 , the
1 electromechanical coupling coefficient k 2 and the f0 Q FoM.
Ref f,C = (11) For a particular material (i.e. fixed k 2 and f0 Q), the lowest
fC
Ref f can be obtained by maximizing C0 . This is achievable
for a capacitor at a frequency f and by connecting multiple piezoelectric layers in parallel, each
!
of which has a poling thickness of d = V /E to sustain
  r r
1 π L C πR
Ref f,LC = tanh = 2π tanh a voltage rating V . For lithium niobate (Table III), we get
f0 C 4Q C L 4
d = 11.5 µm and Ref f = 18 µΩ for a 1 mm3 volume
(12)
assuming optimistically that k 2 = 0.3. This Ref f , normalized
for an LC resonator at the resonance frequency f0 with
by volume, is constant for a particular material. Since vibration
parasitic resistance R [27]. Because a piezoelectric resonator
and poling do not have to be in the same direction (for
can be modeled as a capacitor in parallel with a motional LC
example, vibration along l and poling along d in Fig. 3 (a)),
resonance (Fig. 3 (b)), its effective resistance is
the length of vibration l0 can be varied independent of the
 −1
1 1 poling thickness d. Thus, the low effective resistance of 18 µΩ
Ref f,piezo = + . (13) can theoretically be achieved with 1 mm3 volume of lithium
Ref f,C Ref f, LC
niobate independent of the resonance frequency.
The power output of a ReSC-type circuit with dc bias
However, the above analysis only considers the f0 Q FoM
voltage Vdc and capacitance C at resonance frequency f0 is
and the electrical breakdown of a piezoelectric resonator, and
1  2 2

is an optimistic lower bound on the achievable Ref f . The
P = C (Vdc + Vac ) − (Vdc − Vac ) f0
2 Butterworth-Van Dyke model in (4)–(7) is specific to length-
1 extensional vibration and some material property constants in
= 2CVdc Vac f0 = Vdc Iac , (14)
π Table III are for a specific orientation of the piezoelectric
where Vacpis the peak ac voltage across the capacitor and resonator. The normal mode, with poling and vibration in
Iac = Vac C/L the peak ac current through the capacitor. different directions, may correspond to less favorable material
 6

TABLE IV
R EFERENCE C APACITORS

Dielectric Cref (nF) Vref (V) Vref (mm3 ) u (J/cm3 )

Glass 849.6 300 1 38
C0G 10 250 3.625 86.2 m
X7R 2200 100 4.813 2.29

Fig. 5. Resonator sketches. (a) A multiple-turn inductor for high inductance.

(b) A single-turn inductor for low inductance in parallel with arrays of
reference capacitors (the capacitor self-inductance is not negligible).

property constants and result in a higher effective resis-

tance. Moreover, because the analysis assumes a multilayer
piezoelectric resonator, the poling electrodes contribute to the
motional resistance. The specific mounting position of the
resonator on the base substrate may also affect the kinematic
viscosity η and the FoM. These technological constraints need
to be overcome first to realize the low effective resistance of
the lithium niobate piezoelectric resonator mentioned above.

B. Effective Resistance of an LC Resonator

An electromagnetic resonator as shown in Fig. 5 comprises
a capacitor and an air-core inductor in a total volume V. The Fig. 6. Dissipation factor of C0G and X7R dielectric in optimistic scenarios
volume of the inductor is a particular fraction rv of the total respectively based on Kemet C0402C279K8GAC and C0805C103KCRAC
volume while that of the capacitor equals (1 − rv ) V. For a capacitors [44], and for practical implementations based on the reference
capacitors in Table V, to be discussed in Section VI.
particular volume V, voltage rating V and resonance frequency
f0 , the capacitor dielectric type determines the achievable
energy density and the inductor can be designed to obtain factor. The former can usually be ignored because of the
the desired resonance frequency. We assume a 300 V rating parallel connections among conductor plates. The latter can
to standardize the analysis. be calculated from the dielectric dissipation factor Df , giving
1) Capacitor Model: The achievable energy density of a
Df
dielectric material can be represented by capacitance Cref , RC = . (18)
voltage rating Vref and volume Vref of a reference capacitor 2πf0 C
of that particular dielectric type. For the alkali-free glass di- The dissipation factor of C0G and X7R dielectric materials can
electric discussed in Section II-B, Cref , Vref and Vref can be be extracted from the ESR data of some reference capacitors;
obtained from the dielectric constant and the breakdown field. these reference capacitors are distinct from the capacitors in
For commercially available C0G and X7R capacitors, Digi- Table IV. To obtain approximately the intrinsic loss in the
Key online catalog for surface mount capacitors [41] can be dielectric rather than the conductor plates, these reference
scanned to obtain a capacitor with the highest energy density. capacitors should have a small capacitance and a large voltage
Limiting the voltage rating to lower than 400 V to address a rating in a small case size. We use Kemet K-Sim design
wide range of application, the C0G and X7R capacitors with tool [44] to extract the ESR and Df data (Fig. 6). Table IV and
highest energy density are C2012C0G2E103K125AA from Fig. 6 together give an optimistic combination of high energy
TDK and 12061C225KAT2A from AVX respectively (Ta- density and low dissipation factor; this combination does not
ble IV). It is assumed that these capacitors can be redesigned correspond to any commercially available capacitors.
to obtain the 300 V rating for the analysis. These reference 2) Inductor Model: The inductor needs to be designed such
capacitors, connected in parallel in a volume (1 − rv )V, give that it gives the resonance frequency f0 when combined with
the capacitance the capacitance in (17). This requires
2
Vref (1 − rv )V 1 N 2 Kµ(wL − 4tL /3)(bL − 4tL /3)
C = Cref . (17) L= = , (19)
V2 Vref 4π 2 f02 C lL
For X7R capacitors, the degradation of the dielectric at high where N is the number of turns in the inductor winding and
frequency and operating voltage can be reflected by mul- K Nagaoka’s coefficient, which is the correction factor for
tiplying Cref with a relevant factor. This can be obtained end effects in an air-core coil of finite length [45]. K equals
from experimental measurements or from manufacturer design 1 for narrow and long coils (lL  wL , bL ) and decreases as
tools, namely TDK SEAT2013 [42] and AVX Spicap 3.0 [43]. the coil gets shorter and wider.
The effective series resistance of the capacitor RC can be Low resonance frequencies require a large inductance,
due to conductor plate resistance and dielectric dissipation which, given a constrained volume rv V, can be achieved by
 7

using a multiple-turn winding (N > 1) as shown in Fig. 5 (a).

In this case, the capacitor self-inductance is negligible and the
two sides bL and wL of the inductor are qualitatively the same
so we can set wL = bL . From the required volume rv V, we
can obtain lL = rv V/w2L . Thus, (19) can be rewritten as
2
N 2 Kµ w2L

4tL
L= wL − . (20)
rv V 3
The inductor ESR equals the conductor resistance
N 2 ρe 4(wL − tL ) w2L
RL = , (21)
tL rv V
where ρe is the resistivity of the conductor.
For higher frequencies, only a small inductance is required,
which usually limits the number of turns to one. In this case,
the capacitor self-inductance needs to be considered; a parallel
resonator in this scenario is shown in Fig. 5 (b), which has
  
Kµ0 wL bL 4tL 1 − rv 2tL
L= wL − bL + bL − ,
rv V 3 3rv 3
  
Kµ0 wL bL 4tL 1 + 2rv 2tL
= wL − bL − . (22)
rv V 3 3rv 3
The bL (1 − rv)/(3rv ) term is a first-order approximation Fig. 7. (a) Optimal effective resistance vs. resonance frequency of an LC
resonator and a piezoelectric resonator with 300 V rating and volumes from
for the effective breadth of the capacitors that contribute 1 cm3 to 10−3 mm3 . The optimization assumes an air-core inductor with
to the total inductance. The factor of 3 results from the no skin-effect limit and a lossless capacitor with alkali-free glass dielectric.
assumption that the current is equally distributed among the (b)–(d) Representative inductor designs for each section of the plot.
capacitor electrode plates, resulting in a magnetic field that
increases linearly and a magnetic energy density that increases coil becomes very short, further decreases in the frequency
quadratically along bL , which, when integrated, gives a factor are achieved by increasing the number of turns in the wind-
of 3 in the denominator. The inductor ESR in this case is ings while keeping the dimensions fixed (wL = bL  lL ,
  Fig. 7 (b)). At frequencies higher than the boundary, the
ρe wL bL 1 + 2rv
RL = wL + 2bL . (23) inductors have a single-turn winding, and becomes thinner and
rv V tL 3rv longer (wL = bL  lL , Fig. 7 (d)) as the frequency increases.
3) Resonator Optimization: The effective resistance of the The curve roughness in the higher frequency portion is due
LC resonator in a ReSC circuit can be calculated by substitut- to the accuracy limit in the optimization routine and does not
ing (17)–(23) into (12). Ref f,LC depends on the following: the relate to a physical phenomenon.
resonator volume V, the dielectric type, the inductor-resonator This relation between the minimum effective resistance and
volume ratio rv and the inductor geometry (N , wL , bL and the resonance frequency for a fixed volume can be explained
tL ). According to (19), N can be expressed in terms of wL , as follows. Because the capacitor ESR RC is assumed to
bL and tL . Thus, for a fixed volume V and dielectric type, be zero, design changes along each fixed-volume curve is
the design of the LC resonator depends on four independent mostly due to changes in the inductor geometry. A large
variables (rv , wL , bL , tL ), which can be used to optimize the inductance required for low resonance frequencies is obtained
resonator performance. using an inductor with a multiple-turn winding with a cross-
We perform particle swarm optimization [46] with respect section larger than the length (wL = bL > lL ), resulting
to these parameters to determine the minimum Ref f of an in a large Ref f at the low-frequency end. As the frequency
LC resonator as a function of the resonance frequency for a increases, the inductance is decreased by reducing the number
300 V rating and fixed volumes. Fig. 7 shows the optimization of turns, thereby reducing Ref f . As the number of turns
results for the ideal scenario: an inductor that is not skin-effect approaches one, the self-inductance of capacitors becomes
limited (i.e. using a multilayer foil winding [21] discussed significant, reducing the required lumped inductance and the
in Section II-C) and a lossless capacitor with the high- corresponding inductor ESR, thereby pushing the Ref f vs.
energy-density alkali-free glass dielectric (Table I and IV). f0 curve downwards around the knee in the curve. As the
Each fixed-volume curve has two distinct sections: a larger frequency gets higher, the number of turns remains at one and
slope at the lower frequency end and a smaller slope at the further decreases in L and Ref f are achieved by decreasing the
higher frequency end. The inductor designs for the boundary inductor cross-sectional area (wL , bL ) and increasing its length
between these two sections are a single-turn inductor with lL , resulting in a change in the slope of the curve. Another
an approximately cubic shape (Fig. 7 (c)). As the frequency contribution to the difference between the two sections is the
decreases, the number of turns remain constant at one, and Nagaoka’s coefficient K, which approaches 1 as the inductor
the coil becomes shorter and wider (wL = bL > lL ). As the becomes narrower and longer with increasing frequency.
 8

According to Fig. 7, very low effective resistances can

theoretically be achieved at high frequencies; however, there
are practical limitations regarding the inductor designs. For
example, Ref f as low as 1.32 µΩ can be achieved at 100 MHz
in a 0.1 cm3 volume, but only with impractical dimensions
of 0.22 mm × 0.22 mm × 0.86 m. Because the optimal
inductor is very thin and long, it can also be constructed
by a parallel connection of multiple shorter inductors with-
out significantly affecting the performance; for instance, 200
0.22 mm × 0.22 mm × 4.3 mm in parallel will give the
same Ref f with a more practical aspect ratio. However, im-
plementing the parallel connections usually requires additional
space and results in higher effective resistances. Thus, Fig. 7
shows optimistic physical limits of LC resonator performance
in the ideal scenario of an inductor with no skin-effect limit
and a lossless high-energy-density alkali-free glass dielectric
without the practical constraints regarding aspect ratios or Fig. 8. Optimal effective resistance vs. resonance frequency of an LC
resonator with 300 V rating and 1 cm3 volume. The effects of skin-
implementation of parallel connections. effect limit, C0G and X7R dielectric energy density, and capacitor losses
In order to obtain the performance limit of the LC resonant are examined.
tank rather than that of the ReSC converter, switching losses
are not considered in this analysis. As a result, the effective correspond to distinct reference capacitors. Thus, the “C0G”
resistance gets significantly lower at higher frequencies, but and “X7R”curves in Fig. 8 is an optimistic lower bound on
only with impractical aspect ratios as described above. And achievable Ref f for these capacitor types, calculated using
at MHz switching frequencies, the switching losses may be the highest energy density combined with one of the lower
higher than the effective resistance presented in Fig. 7 even if dissipation factors found in commercial catalogs. For practical
zero-voltage switching (ZVS) is used to minimize losses [47]. implementation, the energy density and the dissipation factor
Thus, very high switching frequencies may not be favorable need to corresponds to the same reference capacitor, which
for designing practical ReSC converters considering all the will invariably increase the achievable Ref f . The results for a
losses, and Fig. 7 is to be regarded as the performance limits practical implementation will be discussed in Section VI.
of the resonators themselves, independent of any assumptions
about semiconductor technology.
C. Effective Resistance Comparison
This optimization process is repeated with various modifica-
tions to determine the effect of various assumptions in the ideal We have modeled and calculated the best possible perfor-
scenario and the results for 1 cm3 volume are plotted in Fig. 8. mance of piezoelectric and LC resonators, with optimistic
We first remove the ideal inductor assumption, i.e. a perfect assumptions, in fixed volumes. Fig. 7 (a) shows the effective
multilayer foil winding that allows full use of a conductor resistance vs. frequency plots of both piezoelectric and LC
much thicker than skin depth. Adding the skin-effect limit resonators of volumes ranging from 1 cm3 to 10−3 mm3 for
increases Ref f by a factor of five at the high-frequency end. comparing the performance of the two resonator types.
Keeping the skin-effect limit, we examine the effect of The effective resistance for the piezoelectric resonators,
changing various assumptions regarding the capacitor. First, calculated using the Butterworth-Van Dyke model in Sec-
the capacitor dielectric is changed from glass to commercial tion III-A is independent of frequency and is inversely propor-
C0G and X7R dielectric materials. The difference in the tional to the total volume. According to Fig. 7, the piezoelec-
energy densities among the three dielectric types translates to tric resonator outperforms the conventional LC resonator in
a difference in achievable Ref f . This is because for the same terms of Ref f below 750 MHz. It should be noted that Fig. 7
resonance frequency, a lower-energy-density dielectric needs shows the optimistic lower bounds on the Ref f at various reso-
a larger inductance than a higher-energy-density dielectric nance frequencies for LC and piezoelectric resonators. Adding
does, and larger inductance usually results in higher ESR as various practical constraints will increase the achievable Ref f ,
discussed above. Second, the degradation of X7R capacitors as shown in Fig. 8 for LC resonators.
with respect to operating voltage and frequency effectively We have analyzed a ReSC-type circuit with a piezoelectric
reduces the capacitor energy density, resulting in a factor of resonator or an LC resonator in terms of Ref f . In the opti-
3.5 increase in Ref f . Third, adding the capacitor loss (18) does mistic scenarios without considering practical implementation
not change Ref f at the low-frequency end but increases Ref f constraints, both resonators have the potential for a low effec-
at the high-frequency end due to the frequency dependence of tive resistance, orders of magnitude better than today’s passive
the capacitor dissipation factor (Fig. 6). The effect of capacitor components. The optimistic assumptions include a perfect
loss is also more pronounced for X7R capacitors than for C0G multilayer piezoelectric resonator without motional resistances
capacitors since the dissipation factor is larger. of the poling electrodes, a capacitor with a high energy density
It is important to emphasize that the dissipation factors dielectric, and a perfect multilayer winding with thin foils and
and the energy densities used in the analysis for Fig. 8 equal current sharing. Thus, these results should be interpreted
 9

as the performance limit of piezoelectric and LC resonators

without considering the current technological constraints.

V. R ESONANT S WITCHED -C APACITOR T YPE C IRCUIT:

P OWER T HROUGHPUT
Because the effective resistances of the resonators are very
small, the more suitable comparison may be in terms of power
throughput. The limitations on power include nonlinearity and
mechanical fracture for the piezoelectric resonator, dielectric
breakdown for the LC resonator, and efficiency and ther-
mal effect of loss for both. Dielectric breakdown is already
considered in terms of voltage rating of reference capacitors
Vref so will not be described explicitly in this section. The
piezoelectric equations in (4)–(10) are only valid for small
linear vibrations and high power may push the resonator
into the nonlinear regime; extreme vibrations may even cause
mechanical breakdown. Efficiency specifications and tolerable Fig. 9. Maximum power output vs. resonance frequency for piezoelectric
thermal effect of loss also limit the current that can be handled. and LC resonators with 300 V rating and 1 mm3 volume. Lithium niobate is
assumed for the piezoelectric resonator. The three sections of the piezo curves
The minimum of these limits sets a constraint on the power with different slopes corresponds to power output limits due to nonlinearity,
handling capability of piezoelectric and LC resonators. the yield strain and the thermal effect of loss. The LC resonator power output
is limited by the thermal effect of loss.

A. Piezoelectric Resonator: Nonlinearity and Yield Strain

To account for nonlinearity in a piezoelectric resonator, the constraint at higher frequencies. We define the lower of these
spring constant can be expanded into higher order terms. For two strain limits as the maximum allowable strain Smax .
a vibration amplitude x, the elastic force acting on an object Since C and L in the Butterworth-Van Dyke model of a
can be expressed as k0 x(1 + k1 x + ...) where k0 is the linear piezolectric resonator (Fig. 3 (b)) corresponds to the stored
term and k1 the first-order anharmonic term of the spring mechanical energy, the maximum vibration amplitude xmax ,
constant. As a condition for unacceptable nonlinearity, we use and hence Smax , can be translated into the maximum ac
the bifurcation point above which a small change in the initial voltage Vac or ac current Iac that can be sustained in the
conditions may cause a jump in the amplitude of the steady resonator. By equating the vibrational energy stored in the
state vibration [48]. The critical vibration amplitude before the piezoelectric resonator at the maximum amplitude k0 x2max /2
vibration becomes nonlinear is given by to the energy stored in the motional inductance at the ac
2
s current limit LIac /2, we get
16
xc = √ 2 , (24) r s
5 3k1 Q C 4k 2 π 2 Y 2 V Smax
Iac = Vac = , (27)
L ρm d(d + 2t) l0
where Q is the quality factor of the resonance [48]. For a bulk
acoustic wave (BAW) resonator as shown in Fig. 3 (a), where d is the poling thickness, t the electrode thickness and l0
π 2 A0 Y 8πν the vibration amplitude. The power output of the piezoelectric
k0 = and k1 = − 0 , (25) resonator can be calculated by substituting (27) into (14).
2l0 3l
Fig. 9 shows the power throughput of a 1 mm3 lithium
where A0 is the cross-sectional area of vibration (bl in niobate piezoelectric resonator with a 300 V rating and dif-
Fig. 3 (a)), l0 the vibration length and ν Poisson’s ratio [48]. ferent values of the f0 Q FoM (1012 , 1013 and 1014 ). The
Combining (24) and (25), we can solve for the maximum strain curve for each value of f0 Q can be divided into three sec-
based on this nonlinearity limit for the piezoelectric resonator, tions. At the low-frequency end, the resonator is limited by
s s
xc 16 3 16 3 p nonlinearity. Because√ the ac current limit Iac is proportional
Sc = 0 = √ = √ f0 . (26) to Smax = Sc ∝ f0 (26) and is inversely proportional to
l 5 3Q 8πν 5 3f0 Q 8πν 3/2
l0 ∝ 1/f0 (8), P ∝ f0 in this nonlinearity-limited section.
Below this strain level, operation can be considered approxi- As the frequency gets higher, the nonlinear strain limit Sc gets
mately linear. Above it, nonlinearity makes operation difficult. larger and the yield strain becomes the binding limit. Because
Operation in the nonlinear range might be feasible, but we find the yield strain is independent of the frequency, the ac current
that there is little motivation to develop this capability, once limit only varies with 1/l0 , and hence P ∝ f0 . Thus, the
other limitations and design considerations are included. slope of the middle section is smaller than that of the first
The other limit on maximum allowable strain is the yield section in Fig. 9. At the high-frequency end, the power output
strain Sy . It can be seen from (26) that for a material with is independent of frequency and is limited by the thermal effect
a particular √value of the f0 Q FoM, the nonlinear strain limit of loss, which will be discussed in Section V-B.
varies with f0 . Thus, Sy rather than Sc becomes the limiting To calculate the power output for each value of the f0 Q
 10

FoM (Fig. 9), we have used the minimum of the nonlinearity, The high power density of the piezoelectric resonator can be
the yield strain and the thermal limits at each frequency. explained by the low Ref f enabled by the perfect multilayer
However, if a higher Ref f is tolerable, the nonlinearity limit resonator we have assumed. Practical considerations such as
can be ignored. Because f0 Q = 1014 (Table III) is the highest the mounting position on the substrate and the poling elec-
achieved with lithium niobate and lower f0 Q products are trodes may introduce additional motional resistances, which
often obtained, it is possible to design a piezoelectric resonator in turn will decrease the thermal limit for the power output.
with any lower f0 Q value. As shown in (16), (26) and (27), a For the LC resonator, one explanation for the favorable
lower f0 Q increases the nonlinearity limited power output, but power output is the high energy density of the capacitor.
only at the expense of a higher Ref f . For instance, the yield The above analyses assume a capacitor with alkali-free glass
strain limited power output at 10 MHz is 2.2 kW whereas the dielectric which has a breakdown electric field of 1.2 V/nm,
nonlinearity limited power output for f0 Q = 1014 is 0.85 kW which allows for a very high energy density of 38 J/cm3
(Fig. 9). Thus, the nonlinear power limit of the piezoelectric (Table I). This is many times higher than the energy density
resonator at 10 MHz may be increased by as much as 2.6 of commercially available capacitors: around 2.3 J/cm3 for
times if we use f0 Q = 1014 /2.62 ≈ 1.5 × 1013 . Thus, by X7R ceramic capacitors and 0.086 J/cm3 for C0G capacitors
varying the f0 Q product for each frequency, the piezoelectric (Table IV). Using these commercially available capacitors
resonator can be designed such that the power output is limited decreases the achievable capacitance, which in turns increases
only by the yield strain and the thermal effect of loss. the required inductance for a particular frequency, resulting in
a higher effective resistance and a lower thermal effect limited
B. Thermal Effect of Loss power output as shown in Fig. 8 and Fig. 9.
The example power circuit we have chosen for the analy-
Power dissipation in the effective resistance of the piezo-
sis, namely the resonant switched-capacitor type circuit, also
electric and the LC resonators increases the temperature of
contributes to the very high power output limit. Because of
the resonators. This thermal effect limits the power loss that
the dc bias, a ReSC-type circuit can be operated with an LC
can be sustained, which in turns limits the power output. The
resonance in which the power processed by the capacitor and
tolerable power loss per unit surface area (Section II-C), and
the inductor may be different by orders of magnitude. For
the resonators’ effective resistance and dimensions can be used
example, for a 1 mm3 LC resonator with an air-core inductor
to calculate the maximum ac current through the resonator, and
and skin-effect limit at 10 MHz, the optimal design has a
hence the thermal effect limited power output.
capacitance of 777 nF, an inductance of 326 pH and a series
As discussed in Section IV-B, long-solenoid inductors can
resistance of 1.3 mΩ, and can sustain an ac peak current of
be redesigned using a parallel connection of multiple shorter
21.3 A. In this design, the inductor only stores a maximum
inductors without impacting the achievable Ref f . Thus, for
energy of 74.1 nJ and processes 0.74 VA at 10 MHz whereas
designs with long-solenoid inductors, it is assumed that the
the capacitor processes as much as 2.04 kVA (14). The fact
inductors are redesigned into an approximately cubic shape.
that such a large difference is possible with a ReSC-type circuit
We use this minimum area to calculate the tolerable winding
is favorable since capacitors have orders of magnitude higher
loss of the inductors in the long-solenoid regime.
energy density than do inductors.
The thermal effect limited power output of the resonators are In non-ReSC-type circuits, such as series resonant convert-
included in Fig. 9 together with other limits discussed above. ers, a small inductance and a large capacitance mean that the
Each “Piezo” curve, which assumes 10 mm2 surface area for loaded quality factor of the circuit will be too small for useful
1 mm3 volume, is limited by the thermal effect of loss at power conversion. For instance, the LC resonator discussed
the high-frequency end. This thermal limit is independent of above only has a characteristic impedance of 20.5 mΩ and
frequency since the optimal Ref f of a piezoelectric resonator an unloaded quality factor of 15.5. To achieve a high loaded
does not depend on the frequency (16). The LC resonators, quality factor for a large load using a series resonant converter,
however, have frequency dependent thermal limits since Ref f the inductance-to-capacitance ratio needs to be much larger.
of the LC resonators varies with the frequency. Because the underlying loss model is also valid for the series
resonant converter, the optimization in Figs. 7, 8 and 9 can
C. Discussion be repeated with a minimum loaded quality factor constraint
Fig. 9 can be used to compare LC and piezoelectric res- for a series resonant converter. This additional constraint may
onators in terms of power throughput. In the most optimistic reduce the achievable power output of the resonator. Thus, the
scenarios for both resonators, the piezoelectric resonator out- ReSC-type circuit plays a significant role in achieving the high
performs the LC resonator both in terms of power throughput power output discussed in this section.
and effective resistance in the frequency range 45–750 MHz.
VI. E XPERIMENTAL R ESULTS
Although piezoelectric resonators have lower Ref f than LC
resonators, they are limited by nonlinearity and yield strain at A. Design Choices
lower frequencies, resulting in a lower power output than the We built a prototype LC resonator to verify the low effective
LC resonators. Nevertheless, both resonators have the potential resistance predicted by Fig. 7 and Fig. 8. However, the results
for power densities that are orders of magnitude higher than in Section IV-B are based on optimistic assumptions; even the
passive components in use today, subject to the caveats noted results for resonators using C0G and X7R capacitors are op-
below. timistic because of the combination of highest energy density
 11

TABLE V
R EFERENCE C APACITORS FOR E XPERIMENTAL V ERIFICATION

Type Manufacturer Part Number Cref (µF) Vref (V) Size (mm × mm × mm) u (J/cm3 )
C0G TDK CGA9N4C0G2E154J230KN 0.15 250 5.7 × 5 × 2.5 65.8 m
X7R AVX 22201C106MAT2A 10 100 5.7 × 5 × 2.79 628.8 m

TABLE VI
E XPECTED AND M EASURED R ESULTS

Measured Measured
Expected
(no core) (core)
L (nH) 1.2 1.48 1.89
C (µF) 2.4 2.35 2.35
f0 (MHz) 2.97 2.70 2.40
RL (mΩ) 0.28 – –
RC (mΩ) 0.16 – –
Resr = RL + RC (mΩ) 0.44 0.56 0.45
Ref f = π 2 Resr /2 (mΩ) 2.17 2.76 2.22
Repr = L/(CResr ) (Ω) 1.07 1.13 1.76

Fig. 10. Minimum achievable Ref f normalized by V 2 vs. resonance

frequency for LC resonators with 1 cm3 volume. The plots include results
for optimistic lower bounds on Ref f of LC resonators using C0G and
X7R capacitors, theoretical values of Ref f using capacitors in Table V and
experimentally measured Ref f for resonators using the C0G capacitor in
Table V.

and low dissipation factor found in online catalogs. Moreover,

the volumes of the reference capacitors used in the analysis Fig. 11. The prototype parallel LC resonator using capacitors in Table V.
(Table IV) are also very small for practical implementation of
a prototype. Thus, for experimental verification, we choose
different reference capacitors with larger volumes (at least frequencies. We chose a design with a predicted resonance
50 mm3 ) that have maximum energy density with voltage frequency of around 3.35 MHz with Ref f ≈ 2.2 mΩ, because
ratings lower than around 3–400 V (Table V). The dissipation the effective resistance is in the low single-digit mΩ and
factors for these reference capacitors are extracted from TDK the lower Ref f at higher frequencies will be much harder
SEAT2013 [42] and AVX Spicap 3.0 [43] software and are to measure. The particular resonator design allocates most of
shown in Fig. 6. More detailed explanations of the capacitor the volume to the capacitors, which means that the capacitor
choice and the small-signal experimental verification can be self-inductance is enough for resonance at around 3 MHz.
found in [12]. The design requires connecting 14 C0G capacitors (Table V)
The energy density and the dissipation factors for these new in a 4 × 3.5 grid, which is impossible using the selected
reference capacitors are used to calculate the effective resis- reference capacitors. Thus, we chose to connect 16 capacitors
tance of an LC resonator with skin-effect limited conductor in a 4 × 4 grid as shown in Fig. 11. This increases the total
that can be practically implemented, and the results are shown volume to 1.14 cm3 , which slightly shifts the “C0G practical”
in Fig. 10. For comparison, the figure also includes Ref f curve in Fig. 10 downwards; however, this shift is negligible
for optimistic resonators with C0G and X7R capacitors from in the order-of-magnitude plot in Fig. 10. Table VI shows the
Section IV-B. Because the voltage ratings for the resonators are expected resonance frequency and various impedances of the
different, Ref f is normalized by V 2 in Fig. 10, which gives a resonator. The table also includes the impedance at resonance
function proportional to loss fraction. For LC resonators using of a parallel resonator with the same inductance, capacitance
C0G and X7R capacitors, the practically achievable Ref f and loss mechanisms, represented by Repr .
and optimistic lower bounds are similar for low frequencies We built a parallel resonator (Fig. 11) considering ease of
(. 100 kHz). However, at higher frequencies, the achievable measuring the 1.07 Ω Repr compared to the 0.44 mΩ Resr .
Ref f are orders of magnitude different, mainly because the A series resonator with the same inductance and capacitance
dissipation factors for off-the-shelf capacitors with high energy can be obtained by removing the copper foil at the top of
density are higher than the optimistic dissipation factors, as the resonator. The measured impedance at resonance of the
shown in Fig. 6. parallel resonator can then be used to calculate the ESR of
We verifed the results for the resonator with C0G capacitors the series resonator, which in turn can be used to calculate
because they have lower ESR than do X7R capacitors at MHz Ref f of the ReSC converter.
 12

B. Small Signal Test

The resulting parallel resonator was characterized using
an Agilent 4294A impedance analyzer. The resonator has a
parallel resonance at 2.70 MHz with a parallel impedance
Repr = 1.13 Ω and a measured capacitance C = 2.35 µF,
resulting in L = 1.48 nH. These measured impedance values
translate to Resr = 0.56 mΩ for the corresponding series
resonator and Ref f = 2.76 mΩ at 2.70 MHz for a ReSC con-
verter using this series resonator, as shown by the asterisk in
Fig. 10. These measured results are also included in Table VI.
This discrepancy between measured and expected f0 and Ref f Fig. 12. The large-signal measurement. (a) The measurement setup showing
may be attributed to the prototype resonator volume which is the impedance-matching transformer, current monitor, the voltage probe points
slightly larger than 1 cm3 , and current crowding at the edges and the parallel resonator. (b) Temperature profile of the parallel resonator
operating around 100 ◦C.
of the capacitor plates. Experimentally, this current crowding
can be reduced by adding blocks of low-loss magnetic cores at
resonator operating near the maximum allowable temperature,
the ends of the resonator to straighten magnetic field lines. We
measured using the thermal camera.
tested this using blocks of Fair-Rite 67 material, which resulted
The drive level was gradually increased until the maximum
in Repr = 1.76 Ω, Resr = 0.45 mΩ and Ref f = 2.22 mΩ at
steady-state surface temperature was around 100 ◦C. Consid-
2.40 MHz (Fig. 10, Table VI). This Ref f with magnetic cores
ering the temperature difference between the surface and the
is closer to the predicted Ref f , verifying the effect of current
inner hotspot, higher drive levels would push the capacitors
crowding. The measured Ref f of the prototype LC resonator
above the rated temperature of 125 ◦C. The drive frequency
matches closely with the calculated Ref f represented by “C0G
had to be adjusted for each drive level and temperature
practical” curve in Fig. 10. This verifies the Ref f model of
so that the measured voltage and current waveforms are in
the LC resonator presented in Section IV-B.
phase, indicating that the resonator is operating at resonance.
The resonance frequency varied from 2.73 MHz at room
temperature to 2.64 MHz at 101 ◦C maximum steady-state
C. Large-Signal Test
surface temperature. This variation in the resonance frequency
A large-signal test was performed to measure the ac voltage is due to the dependence of the dielectric permittivity, hence
capability of the implemented parallel resonator. This ac the capacitance, on the temperature and the drive level.
voltage capability can be used to compute the ac current The impedance at resonance Repr of the parallel resonator
capability of the corresponding series resonator, which in turn was calculated by Repr = Vp,rms /Ip,rms . The ESR of the
can be used to calculate the outpout power of a ReSC resonator corresponding series resonator Resr was then calculated using
using this series resonator. The parallel resonator, rather than Repr , the drive frequency and the inductance measured in the
the series resonator, was used in this large-signal test because small signal test. The measured Repr and the corresponding
of the difficulty of matching a 50 Ω output impedance of a calculated Resr are plotted as a function of the measured
power amplifier to the very low 0.56 mΩ Resr of the series maximum surface temperature in Fig. 13. These measured
resonator; the higher 1.13 Ω Repr of the parallel resonator is resistances at room temperature agree with the small signal test
more easily matched to a 50 Ω impedance. In order to obtain results. At higher drive levels, the higher loss in the resonator
a more conservative estimate of the power handling capability increases the temperature, which results in a lower Repr and a
of the resonator, the large-signal test was performed without higher Resr as shown in Fig. 13. This indicates that the small
using the magnetic cores to straighten the field lines and reduce signal Repr and Resr accurately represent the resonator loss
losses. at room temperature but high-temperature effects may need
The measurement setup (Fig. 12 (a)) includes a 50 Ω to to be considered for a more accurate loss model. However,
3 Ω AVTECH AVX-M4 matching transformer, a Pearson 6585 because the difference between the resistances at 100 ◦C and
current transformer and an Agilent 1141A differential voltage that at the room temperature is only 21%, the presented loss
probe. The measurement setup is based on the configuration model can be considered sufficient in the order-of-magnitude
detailed in [49]. A sinusoidal voltage signal from a function analysis intended for this paper.
generator was amplified by a Tomco BT00500-AlphaS-CW The measured rms voltage of the parallel resonator Vp,rms
power amplifier and fed into the matching transformer. The can be used to calculate the rms current through the corre-
voltage across and the current through the parallel resonator sponding series resonator Is,rms by equating the energy stored
2 2
(Vp and Ip ) were measured using the differential voltage probe in the two resonators CVp,rms = LIs,rms . Fig. 14 shows
and the current transformer respectively. The frequency was Vp,rms and Is,rms as a function of the measured temperature;
tuned to resonance, and at resonance, the power loss in the Is,rms can be as high as 82.4 A for operation at 101 ◦C
resonator was calculated from the product of the rms values of with 25 ◦C ambient temperature. This 82.4 A Irms with
these waveforms. The maximum surface temperature Tmax of 0.67 mΩ Resr gives a total loss of 4.56 W, resulting in a
the resonator was measured using a FLIR E60 thermal camera. loss per unit surface area of 0.6 W/cm2 . This is lower than
Fig. 12 (b) shows the temperature profile of the parallel the 3 W/cm2 tolerable loss limit used in Section II-C to
 13

Fig. 13. Measured impedance at resonance of the parallel resonator and the Fig. 15. Measured power loss of the parallel resonator and the power
corresponding series resonator impedance vs. measured maximum steady-state throughput of a ReSC converter using the corresponding series resonator vs.
surface temperature. measured maximum steady-state surface temperature.

can be used to calculate the power capability of the ReSC

converter; Fig. 15 shows the ReSC converter power output as
well as the corresponding power loss. At high temperatures,
the increase in Pout is smaller because the resonator ESR
increases with the temperature.
It can be seen from Fig. 15 that the presented resonator can
handle a power as high as 7.42 kW with a power loss of only
4.56 W (0.06% loss) attributable to the resonator. This result
is specific to operation at 101 ◦C with an ambient temperature
of 25 ◦C and operation in a higher ambient temperature
will decrease the resonator power capability. However, given
that the resonator only has a 1.14 cm3 volume, it can be
concluded that the power switches and interconnects, rather
than the passive components, limit the performance of a ReSC
converter using this prototype resonator.

Fig. 14. Measured rms voltage of the parallel resonator and the rms current of
the corresponding series resonator vs. measured maximum steady-state surface VII. C ONCLUSION
temperature.
We have presented a fundamental analysis of various energy
calculate the energy density of inductors and in Section V-B storage mechanisms and a possible power circuit for using
to compare the power throughput of the piezoelectric and mass-spring resonance in power converters. Mechanical energy
LC resonators. However, the 3 W/cm2 limit was used as an storage is explored as an alternative to LC resonant tanks
upper limit for loss calculation and a standard for comparing because it provides at least an order of magnitude higher
different resonators rather than as an expected loss; the actual energy density than the electrical alternatives. A comparison
tolerable loss limit depends on the thermal resistance and the of the two types of resonance in the ideal scenarios show
mechanical design of each resonator. that the piezoelectric resonator is more efficient overall and
Consider the application of this resonantor in a simple 1:1 provides a higher power output than LC resonators with skin
ReSC converter. With the 250 V rated capacitors operated effect. However, when used in a resonant switched-capacitor
at 200 V, the input and output voltages of the converter are type circuit, both resonators are capable of very low effective
also 200 V. Although switching losses in presently available resistances and high power handling capabilities in the kW
power devices are likely to be problematic at this voltage range in a 1 cm3 volume, much higher than the capabilities of
and frequency, we assume ideal switch operation in order to the current passive component technology. The low effective
address the capability of the passive component independent of resistance of LC resonators was also verified with a prototype
available semiconductors. Based on the rms current capability resonator which has a low 2.4 mΩ output impedance at
Is,rms , we calculate the dc output current of the ReSC resonance. A large-signal test shows that the resonator is
2
2
√ equating Is,rms Resr =
converter using the series resonator by capable of handling 7.42 kW with only 0.06% loss attributable
Idc Ref f [27], resulting in Idc = 2Is,rms /π; the 82.4 A to the resonator with a maximum surface temperature of
Is,rms results in Idc = 37.1 A. This output voltage and Idc 101 ◦C in an ambient temperature of 25 ◦C.
 14

To examine the future potential, our analysis deliberately [15] Q. Li, K. Han, M. R. Gadinski, G. Zhang, and Q. Wang, “High
omitted practical constraints of today’s technology. This anal- energy and power density capacitors from solution-processed ternary
ferroelectric polymer nanocomposites,” Advanced Materials, vol. 25,
ysis allowed us to eliminate many technologies, but it shows no. 36, pp. 6244–6249, 2014.
that both piezoelectric and electromagnetic resonators may [16] W. Li, L. Jiang, X. Zhang, Y. Shen, and C. Nan, “High-energy-
have the potential to offer much higher efficiency and power density dielectric films based on polyvinylidene fluoride and aromatic
polythiourea for capacitors,” Journal of Materials Chemistry A, vol. 2,
than do typical passive components in use today. The choice no. 38, pp. 15 803–15 807, 2014.
between these two favored technologies is likely to hinge [17] C. Liu, Z. Yu, D. Neff, A. Zhamu, and B. Z. Jang, “Graphene-based
on the factors that have been ignored or idealized in this supercapacitor with an ultrahigh energy density,” Nano Letters, vol. 10,
no. 12, pp. 4863–4868, 2010.
analysis. For the piezoelectric resonator, such factors include [18] A. Yu, V. Chabot, and J. Zhang, Electrochemical supercapacitors for
the specific vibration modes, the mounting type and position, energy storage and delivery: fundamentals and applications. CRC
the poling electrodes and the multilayer structure. For the Press, 2013.
[19] H. Bluhm, Pulsed Power Systems: Principles and Applications. Berlin:
LC resonator, factors such as the dielectric loss and the core Springer, 2006.
loss need to be considered. Because our analysis used high [20] Digi-Key Electronics. Online catalog for surface mount resistors.
performance materials, namely lithium niobate and alkali-free [Online]. Available: http://www.digikey.com/product-search/en/resistors/
chip-resistor-surface-mount/65769 Accessed on: Oct. 27, 2015.
glass dielectric, commercial development of these materials is [21] M. E. Dale and C. R. Sullivan, “General comparison of power loss in
also important for realizing the high performance discussed in single-layer and multi-layer windings,” in IEEE 36th Power Electronics
this paper. The difference between a ReSC-type circuit and a Specialists Conference (PESC). IEEE, 2005, pp. 582–589.
[22] C. R. Sullivan, D. V. Harburg, J. Qiu, C. G. Levey, and D. Yao, “Inte-
series resonant converter, and hence the effect of circuit design, grating magnetics for on-chip power: A perspective,” IEEE Transactions
also needs to be explored. The effects of all of these factors on Power Electronics, vol. 28, no. 9, pp. 4342–4353, 2013.
are important topics for future research. [23] K. Uchino, “Piezoelectric ultrasonic motors: overview,” Smart Materials
and Structures, vol. 7, no. 3, p. 273, 1998.
[24] B. Koc, S. Cagatay, and K. Uchino, “A piezoelectric motor using two
R EFERENCES orthogonal bending modes of a hollow cylinder,” IEEE Transactions on
[1] C. R. Sullivan, D. Yao, G. Gamache, A. Latham, and J. Qiu, “(Invited) Ultrasonics, Ferroelectrics and Frequency Control, vol. 49, no. 4, pp.
Passive component technologies for advanced power conversion enabled 495–500, 2002.
by wide-band-gap power devices,” ECS Transactions, vol. 41, no. 8, pp. [25] Y. Liu, W. Chen, P. Feng, and J. Liu, “A rotary piezoelectric motor
315–330, 2011. using bending vibrators,” Sensors and Actuators A: Physical, vol. 196,
[2] J. W. Kolar, D. Bortis, and D. Neumayr, “The ideal switch is not pp. 48–54, 2013.
enough,” in Power Semiconductor Devices and ICs (ISPSD), 28th [26] M. Shoyama, T. Naka, and T. Ninomiya, “Resonant switched capacitor
International Symposium on. IEEE, 2016, pp. 15–22. converter with high efficiency,” in IEEE 35th Annual Power Electronics
[3] D. J. Perreault, J. Hu, J. M. Rivas, Y. Han, O. Leitermann, R. C. N. Specialists Conference (PESC), vol. 5. IEEE, 2004, pp. 3780–3786.
Pilawa-Podgurski, A. Sagneri, and C. R. Sullivan, “Opportunities and [27] K. Kesarwani and J. T. Stauth, “A comparative theoretical analysis of
challenges in very high frequency power conversion,” in 24th Annual distributed ladder converters for sub-module PV energy optimization,”
IEEE Applied Power Electronics Conference and Exposition (APEC), in IEEE 13th Workshop on Control and Modeling for Power Electronics
2009, pp. 1–14. (COMPEL), 2012, pp. 1–6.
[4] C. Ó. Mathúna, N. Wang, S. Kulkarni, and S. Roy, “Review of integrated [28] K. S. Van Dyke, “The piezo-electric resonator and its equivalent net-
magnetics for power supply on chip (PwrSoC),” IEEE Transactions on work,” Proceedings of the Institute of Radio Engineers, vol. 16, no. 6,
Power Electronics, vol. 27, no. 11, pp. 4799–4816, 2012. pp. 742–764, 1928.
[5] J. M. Noworolski and S. R. Sanders, “An electrostatic microresonant [29] M. Pijolat, A. Reinhardt, E. Defay, C. Deguet, D. Mercier, M. Aid,
power conversion device,” in 23rd Annual IEEE Power Electronics J. S. Moulet, B. Ghyseln, D. Gachon, and S. Ballandras, “Large Qxf
Specialists Conference (PESC). IEEE, 1992, pp. 997–1002. product for HBAR using smart cutTM transfer of LiNbO3 thin layers
[6] C. A. Rosen, “Analysis and design of ceramic transformers and filter onto LiNbO3 substrate,” in IEEE Ultrasonic Symposium (IUS). IEEE,
elements,” Ph.D. dissertation, Syracuse University, 1956. 2008, pp. 201–204.
[7] Y. Fuda, K. Kumasaka, M. Katsuno, H. Sato, and Y. Ino, “Piezoelec- [30] S. A. Chandorkar, M. Agarwal, R. Melamud, R. N. Candler, K. E.
tric transformer for cold cathode fluorescent lamp inverter,” Japanese Goodson, and T. W. Kenny, “Limits of quality factor in bulk-mode
Journal of Applied Physics, vol. 36, no. 5S, pp. 3050–3052, 1997. micromechanical resonators,” in IEEE 21st International Conference on
[8] J. Yoo, K. Yoon, S. Hwang, S. Suh, J. Kim, and C. Yoo, “Electrical Micro Electro Mechanical Systems (MEMS). IEEE, 2008, pp. 74–77.
characteristics of high power piezoelectric transformer for 28 W fluo- [31] S. Gong and G. Piazza, “Design and analysis of lithium–niobate-based
rescent lamp,” Sensors and Actuators A: Physical, vol. 90, no. 1, pp. high electromechanical coupling RF-MEMS resonators for wideband
132–137, 2001. filtering,” IEEE Transactions on Microwave Theory and Techniques,
[9] T. Zaitsu, Y. Fuda, Y. Okabe, T. Ninomiya, S. Hamamura, and M. Kat- vol. 61, no. 1, pp. 403–414, 2013.
suno, “New piezoelectric transformer converter for ac-adapter,” in [32] Boston Piezo Optics Inc. Lithium niobate physical properties. [Online].
Twelfth Annual Applied Power Electronics Conference and Exposition Available: http://www.bostonpiezooptics.com/lithium-niobate Accessed
(APEC), vol. 2. IEEE, 1997, pp. 568–572. on: Mar 25, 2015.
[10] O. Ohnishi, Y. Sasaki, T. Zaitsu, H. Kishie, and T. Inoue, “Piezoelectric [33] A. Rivera, G. Garcia, J. Olivares, M. L. Crespillo, and F. Agulló-
ceramic transformer for power supply operating in thickness extensional López, “Elastic (stress-strain) halo associatd with ion-induced nano-
vibration mode,” IEICE transactions on fundamentals of electronics, tracks in lithium niobate: role of crystal anisotropy,” Journal of Physics
communications and computer sciences, vol. 77, no. 12, pp. 2098–2105, D: Applied Physics, vol. 44, no. 47, p. 475301, 2011.
1994. [34] M.-J. Jin, O.-Y. Jeon, B.-J. Kim, and M. Cha, “Fabrication of pe-
[11] P. A. Kyaw and C. R. Sullivan, “Fundamental examination of multiple riodically poled lithium niobate crystal and poling-quality evaluation
potential passive component technologies for future power electronics,” by diffraction measurement,” Journal of the Korean Physical Society,
in IEEE 16th Workshop on Control and Modeling for Power Electronics vol. 47, pp. S336–S339, 2005.
(COMPEL), 2015. [35] M. Domenjoud, M. Lematre, M. Gratton, M. Lethiecq, and L.-P. Tran-
[12] P. A. Kyaw, A. L. F. Stein, and C. R. Sullivan, “Optimizing resonant Huu-Hue, “Theoretical and experimental study of the electroacoustic
tanks for high power density using standard capacitors,” in IEEE 18th behavior of lithium niobate under an initial mechanical stress,” IEEE
Workshop on Control and Modeling for Power Electronics, 2017. Transactions on Ultrasonics, Ferroelectrics and Frequency Control,
[13] M. F. Ashby, Materials selection in mechanical design, 4th ed. Burling- vol. 60, no. 10, pp. 2219–2224, 2013.
ton, MA: Butterworth-Heinemann, 2011. [36] C. Schaef and J. T. Stauth, “A 3-phase resonant switched capacitor
[14] N. J. Smith, B. Rangarajan, M. T. Lanagan, and C. G. Pantano, “Alkali- converter delivering 7.7 W at 85% efficiency using 1.1 nH PCB trace
free glass as a high energy density dielectic material,” Materials Letters, inductors,” IEEE Journal of Solid-State Circuits, vol. 50, no. 12, pp.
vol. 63, no. 15, pp. 1245–1248, 2009. 2861–2869, 2015.
 15

[37] S. Pasternak, C. Schaef, and J. Stauth, “Equivalent resistance approach Phyo Aung Kyaw (S’15)received the B.A. degree in
to optimization, analysis and comparison of hybrid/resonant switched- physics from the Amherst College, Amherst, MA,
capacitor converters,” in Control and Modeling for Power Electronics USA, in 2014. He is currently working towards
(COMPEL), IEEE 17th Workshop on, 2016, pp. 1–8. his Ph.D. degree in electrical engineering at Dart-
[38] A. Blumenfeld, A. Cervera, and M. M. Peretz, “Enhanced differential mouth College, Hanover, NH, USA. His research
power processor for pv systems: Resonant switched-capacitor gyrator interests include power electronics and magnetics,
converter with local mppt,” IEEE Journal of Emerging and Selected electromagnetic and piezoelectric resonators, high-
Topics in Power Electronics, vol. 2, no. 4, pp. 883–892, 2014. frequency passive components, and wireless power
[39] K. Kesarwani and J. T. Stauth, “Resonant and multi-mode operation of transfer.
flying capacitor multi-level dc-dc converters,” in Control and Modeling
for Power Electronics (COMPEL), IEEE 16th Workshop on, 2015, pp.
1–8.
[40] Y. Li, J. Chen, M. John, R. Liou, and S. R. Sanders, “Resonant switched Aaron L.F. Stein received the Ph.D. degree in
capacitor stacked topology enabling high dc-dc voltage conversion ratios electrical engineering and computer science from the
and efficient wide range regulation,” in Energy Conversion Congress and University of Michigan, Ann Arbor, MI, USA, in
Exposition (ECCE), IEEE, 2016. 2016, and is currently a Post-Doctoral Research As-
[41] Digi-Key Electronics. Online catalog for surface mount ceramic sociate at Dartmouth College, Hanover, NH, USA.
capacitors. [Online]. Available: http://www.digikey.com/products/en/ He has published technical papers on topics includ-
capacitors/ceramic-capacitors/60 Accessed on: Mar 06, 2017. ing wireless power transfer, energy harvesting, and
[42] TDK. (2015) SEAT 2013 - SElection Assistant of TDK components. electromagnetic components.
[Online]. Available: https://product.tdk.com/info/en/technicalsupport/
seat/index.html Accessed on: Feb 06, 2017.
[43] AVX. (2015) Spicap 3.0. [Online]. Available: http://www.avx.com/
resources/design-tools/ Accessed on: Oct. 03, 2015.
[44] Kemet. K-sim (webspice). [Online]. Available: http://ksim.kemet.com/ Charles R. Sullivan (S’93-M’96-SM’12-F’14) re-
Ceramic/CeramicCapSelection.aspx Accessed on: Nov 08, 2016. ceived the B.S.(Hons.) degree in electrical engineer-
[45] “Radio instruments and measurements,” Bureau of Standards Circular, ing from Princeton University, Princeton, NJ, USA,
vol. C74, p. 252, January 1937. in 1987, and the Ph.D. degree in electrical engineer-
[46] S. Chen. Particle swarm toolbox for MATLAB. [Online]. Available: ing from the University of California, Berkeley, CA,
https://code.google.com/p/psomatlab/ Accessed on: Feb 19, 2015. USA, in 1996.
[47] J. B. Fedison, M. Fornage, M. J. Harrison, and D. R. Zimmanck, “Coss Between the B.S. and Ph.D. degrees, he was
related energy loss in power MOSFETs used in zero-voltage-switched with Lutron Electronics designing electronic bal-
applications,” in Applied Power Electronics Conference and Exposition lasts. He is currently a Professor at Thayer School
(APEC), Twenty-Ninth Annual IEEE, 2014, pp. 150–156. of Engineering at Dartmouth, Hanover, NH, USA.
[48] V. Kaajakari, T. Mattila, A. Oja, and H. Seppä, “Nonlinear limits for His research interests include design optimization of
single-crystal silicon microresonators,” Journal of Microelectromechan- magnetics for power applications, energy efficiency, and renewable energy,
ical Systems, vol. 13, no. 5, pp. 715–724, 2004. and electromagnetic modeling of capacitors.
[49] H. Syed and C. R. Sullivan, “High frequency magnetic toroidal core loss Dr. Sullivan received the National Science Foundation CAREER Award
measurement fixture,” in 28th Annual IEEE Applied Power Electronics and two Power Electronic Society Prize Paper Awards.
Conference and Exposition (APEC), 2013, pp. 2466–2473.