SURVEY OF STATE-OF-THE-ART TECHNOLOGIES FOR … server... · 2016-09-26 · December 10, 2004 10:0...
Transcript of SURVEY OF STATE-OF-THE-ART TECHNOLOGIES FOR … server... · 2016-09-26 · December 10, 2004 10:0...
December 10, 2004 10:0 WSPC/123-JCSC 00165
Journal of Circuits, Systems, and ComputersVol. 13, No. 3 (2004) 399–423c© World Scientific Publishing Company
SURVEY OF STATE-OF-THE-ART TECHNOLOGIES FOR
COMPUTER NETWORK SERVER AC/DC POWER SUPPLIES
MILAN M. JOVANOVIC
Power Electronics Laboratory,
Delta Products Corporation, P. O. Box 12173,
5101 Davis Drive, Research Triangle Park, NC 27709, USA
Due to a continuously increasing power demand of computer–network server equipment,power supply manufacturers are faced with requirements for ac/dc server power supplieswith significantly higher power densities. In this paper, circuit technologies and optimiza-tion techniques that are suitable for meeting these new power-density requirements arediscussed. The focus of the paper is on single-phase, single-output ac/dc power suppliesthat are used to generate a dc-bus voltage in distributed power systems employed inhigh-end network servers and telecom systems.
Keywords: ac/dc converters; off-line converters; PFC; boost converters; full-bridge con-verters; soft-switched converters; zero-voltage-switching (ZVS); hold-up-time.
1. Introduction
Recently, high power density has become one of the major requirements in today’s
off-line power converters for computer–network servers and telecom equipment pri-
marily due to a continuous growth in their system power needs. Some projections
put the expected annual growth of power density of off-line power supplies at 10–
15%.1 Specifically, in the near future, multiple-output ac/dc server power supplies
are expected to attain power densities of over 7–10 W/in3, whereas ac/dc front-
end power supplies for distributed power systems (DPSs) employed in high-end
servers and telecom applications to generate a 12-V, 24-V, or 48-V dc-bus voltage
are expected to exhibit power densities in the 15–20 W/in3 range. This already
challenging task of continuous power-density improvements is further compounded
by low profile (height) requirements of today’s server/telecom ac/dc power supplies,
which adversely affects the optimal selection of some power components. Typically,
the profile of today’s state-of-the-art server/telecom power supplies is 1U or 2U,
where the height of 1U (unit) is equal to 1.75 inches.
Generally, to achieve the expected power densities, it is necessary to optimize
power-stage performance and use advanced packaging/thermal management tech-
niques. Namely, the power-stage performance optimization requires the employment
of efficient high-frequency topologies that can reduce the size of the power stage
399
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
400 M. M. Jovanovic
PFC FRONT END
RECTIFIER &
EMI FILTER DC/DC OUTPUT STAGE
VOVIN
HOLD-UP-TIME
CAPACITOR(S)
Fig. 1. Block diagram of typical ac/dc power supply.
without sacrificing its efficiency. This typically requires a soft-switching topology
with optimized magnetic and semiconductor components so that the switching and
conduction losses are kept at a minimum. However, an efficient power stage alone
is not sufficient to bring about a significant power-density improvement. In fact, in
today’s ac/dc power supply, the major opportunity for substantial power-density
increases lays in the replacement of old, conventional packaging and thermal man-
agement techniques with new ones that would employ high-level component inte-
grations, as well as new and more suitable materials. Currently, the packaging and
thermal issues have been addressed in depth by researchers at National Science
Foundation’s Center for Power Electronics Systems (CPES) at Virginia Tech.2–4
This paper does not deal with packaging and thermal management issues and tech-
niques, but only with electrical circuit design issues related to high-power-density
server power supplies.
As shown in Fig. 1, a typical ac/dc server power supply power stage consists
of four major functional parts that require a significant volume. Specifically, the
EMI filter, power-factor-correction (PFC) circuit, hold-up time energy-storage ca-
pacitor(s), and dc/dc output stage are integral parts of every ac/dc server power
supply. The telecom ac/dc power supplies have the same structure except that the
energy-storage capacitor(s) is sized based on the PFC circuit output ripple rather
than based on the hold-up time specifications. To increase the power density, it is
necessary to optimize the performance and minimize the size of all four parts since
a partial optimization is insufficient to meet the future power-density requirements.
In this paper, circuit technologies and design optimization techniques suitable
for high-density ac/dc server/telecom power supplies are reviewed. Specifically,
technologies for high-performance PFC and dc/dc circuits, as well as techniques
for hold-up time capacitor minimization, are discussed in details. Due to the ab-
sence of a practical EMI-filter design procedure, which is direct result of a lack of
a systematic research effort in the EMI optimization area, the major EMI design
issues are only briefly addressed. In addition, the focus of the present paper is on
technologies for single-output ac/dc power supplies that are used as front ends in
high-end server/telecom DPSs, i.e., on power-supplies with a 48-V or 12-V single
output, which are expected to exhibit power densities over 15 W/in3.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 401
2. PFC Technologies
At higher power levels, the continuous-conduction-mode (CCM) boost converter is
the preferred topology for implementation of a front-end with active input-current
shaping. As a result, in recent years, significant efforts have been made to improve
the performance of high-power boost converters. The majority of these development
efforts have been focused on reducing the adverse effects of the reverse-recovery
characteristic of the boost rectifier on the conversion efficiency and electromagnetic
compatibility (EMC).5
Generally, the reduction of reverse-recovery-related losses and EMC problems
requires that the boost rectifier be “softly” switched off by controlling the turn-off
rate of its current. So far, a number of soft-switched boost converters and their
variations have been proposed.6–23 All of them use additional components to form
a passive snubber,6–11 or an active snubber12–23 circuit that controls the turn-off
di/dt rate of the boost rectifier. Generally, the passive snubber approaches use only
passive components such as resistors, capacitors, inductors, and rectifiers, whereas
active snubber approaches also employ active switch (es).
Although passive lossless snubbers can improve efficiency, their performance
is not good enough to make them viable candidates for applications in high-
performance PFC circuits. Namely, the reported efficiency improvements for various
passive lossless snubbers is in the 1–2% range depending on the line and load con-
ditions, switching frequency, and boost rectifier characteristic.7,8,10 However, they
suffer from increased component stresses and are not able to operate with soft
switching of the boost switch, which is detrimental in high-density applications
that require increased switching frequencies.
The simultaneous reduction of the reverse-recovery losses and soft switching
can be achieved by active snubbers. As an example, Fig. 2 shows a boost converter
CR
CF
D
LR
VIN
S1
VO
+
-D1
S
L
RL
Fig. 2. Boost PFC power stage with active snubber.12,13
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
402 M. M. Jovanovic
with an active snubber that employs a snubber inductor connected to the common
node of the boost switch and the rectifier to control the turn-off rate of the rectifier
current.12,13 In the circuit in Fig. 2, boost switch S turns on with zero voltage
switching (ZVS) and rectifier D turns off under soft switching conditions. However,
auxiliary switch S1 is turned on while its voltage is equal to the output voltage,
and subsequently turned off while carrying a current greater than the input current.
Although it is switched under “hard” switching conditions, the power dissipation of
the auxiliary switch in a properly designed converter is much smaller than that of
the main switch because the auxiliary switch conducts only during switching transi-
tions. The major drawback of the soft-switched converter in Fig. 2 is the undesirable
resonance of output capacitance COSS1 of the auxiliary switch (not shown in Fig. 2)
and snubber inductor LR that occurs when auxiliary switch S1 is opened and after
the current in LR falls to zero. This parasitic resonance generates an undesirable
current through LR, which increases the conduction losses and upsets the normal
operation of the circuit. To reduce the effect of this resonance, it is necessary to add
a rectifier and a saturable inductor in series with LR,13 which somewhat degrades
the conversion efficiency and increases the component count and cost of the circuit.
The performance of the circuit in Fig. 2 can be improved by implementing
an active snubber with soft switching of the auxiliary switch. Figure 3 shows the
implementation of the active snubber that beside ZVS of the main switch and
soft switching of the boost rectifier, offers zero current switching (ZCS) of the
auxiliary switch.14 Another implementation of the active snubber that offers ZVS
of the main switch and ZCS of the auxiliary switch is shown in Fig. 4.15 In both
implementations, a transformer, connected to the output of the boost converter, is
used to provide voltage across resonant inductor LR to decrease the current through
CR
CF
D
LR
VIN
S1
VO
+
-
D1
S
L
RL
D2
N1
N2
Fig. 3. Boost PFC power stage with active snubber.14
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 403
CR
CF
D
LR
VIN
VO
+
-
S1
S
L
RL
TR
1:n
Fig. 4. Boost PFC power stage with active snubber.15
FV
IN
S
S1
VO
+
-C
LS
D1
DL
C RL
C
Fig. 5. Boost PFC power stage with active snubber.16
the inductor and, therefore, auxiliary switch S1, down to zero before the switched is
turned off. The circuits in Figs. 3 and 4 are expected to show similar performance.
Soft switching of all semiconductor components can also be achieved by employ-
ing an active snubber implemented with snubber inductor LS in series with main
switch S and rectifier D, as shown in Figs. 5 and 6.16,17 Both circuits offer ZVS
turn-on of the main and the auxiliary switches, and ZVS turn-off of the rectifier. In
addition, the circuits in Figs. 5 and 6 do not suffer from parasitic resonance, and,
therefore, do not require additional components. However, the voltage stress of the
main switch in these implementations is higher than that in the circuits in Figs. 2
through 4 because of the placement of the snubber inductor in series with the main
switch and the rectifier. Nevertheless this increased voltage stress can be minimized
by a proper selection of circuit components and the switching frequency.16–18
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
404 M. M. Jovanovic
CF
L D
LS
VIN
S1
VC+
VO
+
-
CC
-
RLDC
S
Fig. 6. Boost PFC power stage with active snubber.17
The difference between the implementations in Figs. 5 and 6 is in the point of
connection of the active snubber consisting of a series connection of auxiliary switch
S1 and clamp capacitor CC . In the circuit in Fig. 5, the active snubber is connected
to the anode of rectifier D, whereas in the circuit in Fig. 6 it is connected to the
cathode of rectifier D. As a result, the circuits have different current waveshapes
through the components and require different timing of the drive signals for the
main and auxiliary switches. Specifically, the circuit in Fig. 5 requires a more desir-
able overlapping gate drive, whereas the circuit in Fig. 6 requires a nonoverlapping
gate drive that is, generally, more susceptible to noise than an overlapping gate
drive. It should be noted that both circuits require a high-side driver for the aux-
iliary switch. However, the circuit in Fig. 6 can be implemented with an isolated
active snubber, which has both the main and the auxiliary switches referenced
to ground so that a direct (low-side), more desirable drive can be used for both
switches.19
Due to ZVS of the boost switch, the efficiency of the circuits in Refs. 12–19
was shown to be significantly improved when the boost switch is implemented with
a MOSFET (metal-oxide-semiconductor field-effect transistor) device. Generally,
the circuits in Fig. 2 through 6 improve the low-line conversion efficiency of a
universal-line boost PFC for approximately 3–4%.12–19 However, the improvement
in the efficiency of these circuits when used with an IGBT (insulate-gate bipolar
transistor) device is expected to be much less because for IGBT devices zero-current
switching (ZCS) is the optimal switching strategy.20
A ZCS boost converter suitable for applications with IGBTs was introduced in
Ref. 21. Although in this circuit the boost switch is turned off with zero current,
the circuit exhibits an undesirable resonance between the snubber inductor and the
output capacitance of the switches, which requires additional clamp and/or snubber
circuits.22 The required clamp and snubber circuits not only increase the complexity
and cost of the circuit, but also have a detrimental effect on its efficiency.
A soft-switching technique which is suitable for IGBT applications, and which
does not suffer from undesirable resonance of circuit’s components, was introduced
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 405
L DL S
VIN VO
+
-
RLCDC
CC
S1S
F
Fig. 7. Boost PFC power stage with active snubber suitable for IGBT implementations.23
in Ref. 23. This technique improves the performance of the PFC boost converter
by eliminating the switching losses with a new zero-current-zero-voltage-switched
(ZC-ZVS) active-snubber circuit that consists of a snubber inductor, a clamp diode,
a clamp capacitor, and an auxiliary switch, as shown in Fig. 7. The ZC-ZVS snub-
ber reduces the reverse-recovery-related losses of the rectifier and also provides
soft switching of the main and auxiliary switches. Specifically, the main switch
turns off with ZCS, whereas the auxiliary switch turns on with ZVS. In addition,
because the proper operation of the ZC-ZVS snubber circuit requires that the con-
duction period of the main and the auxiliary switches overlap, the circuit in Fig. 7
is not susceptible to failures due to accidental overlapping of the main and auxil-
iary switch gate drives. Furthermore, the complexity and cost of the converter are
further reduced because the converter requires simple nonisolated gate drives for
both switches.
Finally, it should be noted that the recent advancements in the development
of silicon carbide (SiC) rectifiers that have made SiC rectifiers commercially avail-
able are dramatically changing the boost converter design-optimization priorities.24
Since SiC rectifiers virtually have no reverse-recovery charge, the reverse-recovery
related losses are not an issue and, consequently, a snubber circuit that controls the
turn-off rate of the rectifier current is not needed anymore. In fact, SiC technology
makes it feasible to dramatically increase the switching frequency of the boost con-
verter and, eventually, reduce the size of the circuit, primarily the size of the boost
inductor.25 However, the dramatic increase of the switching frequency without a
deterioration of efficiency is only possible if the boost switch is soft-switched. As a
result, a high-frequency boost PFC with a SiC rectifier still requires additional cir-
cuitry which is employed to create soft-switching conditions. Therefore, for PFCs
with SiC rectifiers, the optimization focus is shifted from the reduction of recti-
fier switching losses to the minimization of switching losses of the boost switch.
This optimization shift is likely to bring about a variety of new implementations of
soft-switched boost PFC circuits.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
406 M. M. Jovanovic
LLK
CB
VIN
Q 1
Q 2
Q 3
Q 4
C1
D1
C2
D2
C3
D3
C4
D4
TR
o V
D1
D2
NP
NS
NS
L F
CF RL
I O
Fig. 8. Conventional FB ZVS-PWM dc/dc converter.
3. Technologies for DC/DC Output Stage
The soft-switched, full-bridge (FB) PWM converter shown in Fig. 8 is the most
widely used circuit in front-end ac/dc power supplies for server and telecom
DPSs.26–28 This converter features ZVS of the primary switches with a relatively
small circulating energy and at a constant switching frequency. The control of the
output voltage at a constant frequency is achieved by the phase-shift technique. In
this technique, the turn-on of a switch in the Q3–Q4 leg of the bridge is delayed, i.e.,
phase-shifted, with respect to the turn-on instant of the corresponding switch in the
Q1–Q2 leg. The ZVS of the lagging-leg switches Q3 and Q4 is achieved primarily
by the energy stored in output filter inductor LF . Since the inductance of LF is
relatively large, the energy stored in LF is sufficient to discharge output parasitic
capacitance C3 and C4 of switches Q3 and Q4 in the lagging-leg and to achieve ZVS
even at very light load currents. However, the discharge of parasitic capacitance C1
and C2 of leading-leg switches Q1 and Q2 is done by the energy stored in leakage
inductance LLK of the transformer because during the switching of Q1 or Q2, the
transformer primary is shorted by the simultaneous conduction of rectifiers D1 and
D2 that carry the output filter inductor current. Since leakage inductance LLK is
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 407
small, the energy stored in LLK is also small so that ZVS of Q1 and Q2 cannot be
achieved even at relatively high output currents. The ZVS range of the leading-leg
switches can be extended to lower load currents by intentionally increasing the leak-
age inductance of the transformer and/or by adding a large external inductance in
series with the primary of the transformer. If properly sized, the external inductance
can store enough energy to achieve ZVS of the leading-leg switches even at very low
currents. However, a large external inductance also stores extremely high energy
at full load that produces a large circulating energy, which adversely affects the
stress of the semiconductor components and the conversion efficiency. In addition,
a large inductance in series with the primary winding (leakage inductance and/or
external inductance) reduces the effective duty cycle on the secondary side because
it extends the time required to change polarity of the primary current. To make
up for the reduced duty cycle, the turns ratio of the transformer must be reduced,
which has detrimental effect on the performance because of increased component
losses and stresses.27
Another major limitations of the circuit in Fig. 8 is the severe parasitic ring-
ing at the secondary of the transformer during the turn-off of a rectifier caused
by the resonance of the rectifier’s junction capacitance with the leakage induc-
tance of the transformer and/or the external inductance. To control the ringing, a
heavy snubber circuit needs to be used on the secondary side, which may signifi-
cantly lower the conversion efficiency of the circuit.28 In fact, the loss of duty cycle
and secondary-side parasitic ringing are the major factors that limit the maximum
switching frequency of kilowatt-range FB ZVS-PWM circuits to around 100 kHz.29
The ZVS range of the leading-leg switches in the FB ZVS-PWM converter in
Fig. 8 can be extended to lower load currents without a significant increase of
the circulating energy by using a saturable external inductor instead of a linear
inductor.29 If the saturable inductor is designed so that it saturates at higher load
currents, it will not store excessive energy at high loads. At the same time, at low
load currents when the inductor is not saturated, it will have a sufficiently high
inductance to store enough energy to provide ZVS of the leading-leg switches even
at very light loads. While it was demonstrated that a properly designed saturable
inductor can improve the performance of the FB ZVS-PWM converter, the circuit
requires a relatively large-size magnetic core to implement the inductor, which
increases the cost of the circuit. Generally, a larger core is required to eliminate the
thermal problem that is created by excessive core loss because the saturable core is
placed in the primary circuit and its flux swings between the positive and negative
saturation levels.
The ZVS range of the FB ZVS-PWM converter can also be extended to lower
load currents by placing saturable cores on the secondary side, as shown in Fig. 9.
Since in the implementation in Fig. 9, saturable cores LSAT1 and LSAT2 are con-
nected in series with rectifiers D1 and D2, their flux swing is confined between zero
and the positive saturation level, i.e., it is a half of the flux swing of the saturable
core in Fig. 9. As a result, the total core loss in the circuit in Fig. 9 is reduced
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
408 M. M. Jovanovic
LSAT1
LSAT2
CB
VIN
Q 1
Q 2
Q 3
Q 4
C1
D1
C2
D2
C3
D3
C4
D4
TR
o V
D1
D2
NP
NS
NS
L F
CF RL
I O
DFW
Fig. 9. FB ZVS-PWM converter with secondary-side saturable inductors.30
compared to that of the circuit in Fig. 8. However, because in voltage step-down
converters (i.e., converters with output voltage VO smaller than input voltage VIN),
secondary currents are larger than the primary current, the total copper loss of the
saturable core windings in Fig. 9 is increased compared to the corresponding loss
in the circuit in Fig. 8. An additional feature of secondary-side saturable inductors
LSAT1 and LSAT2 is that they serve as turn-off snubbers for rectifiers D1 and D2,
damping the parasitic oscillations between the junction capacitance of the rectifiers
and the leakage inductance of the transformer, as well as helping to reduce the
reverse-recovery current losses if fast-rectifier rectifiers are used.
The FB ZVS-PWM with secondary-side saturable inductors can be implemented
with or without freewheeling rectifier DFW , which is shown with dashed lines in
Fig. 9. The operation of the circuit in Fig. 9 with freewheeling diode DFW is identi-
cal to that with the saturable inductor on the primary side.29 Namely, with DFW ,
saturable inductors LSAT1 and LSAT2 are used to store enough energy at lower load
currents so that ZVS of the primary switches is achieved with minimum circulating
energy. However, in the implementation without freewheeling diode DFW , saturable
inductors LSAT1 and LSAT2 are not used for energy storage.30,31 Instead, they are
used to briefly delay the conduction of the nonconducting rectifier after a switch in
a bridge leg is turned off and, consequently, force the current through filter inductor
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 409
LF to continue to flow through the rectifier which already has been conducting. As
a result, in the implementation of the circuit in Fig. 9 without freewheeling rectifier
DFW , the energy stored in output filter inductor LF is used to create ZVS condi-
tion for leading-leg switches Q1 and Q2 in the same way as it is used to create ZVS
conditions for switches Q3 and Q4 in the lagging-leg, i.e., by utilizing the energy
stored in output-filter inductor LF .
An approach that deals with the secondary-side ringing problem in a simple
and effective way was introduced in Ref. 32. In this approach, shown in Fig. 10,
the secondary-side ringing due to the resonance of the capacitance of the secondary-
side rectifier and external primary inductor that is employed to extend the ZVS
range is controlled by primary-side clamp diodes DC1 and DC2. If the leakage in-
ductance of the transformer is minimized, the primary clamp circuit also effectively
clamps the voltage across the secondary winding so that parasitic ringing and its
associated voltage stress are virtually eliminated. Due to its simplicity, the circuit in
Fig. 10 has found use in many applications, particularly, in high-power switch-mode
rectifiers (SMRs) employed in telecommunication power systems. However, the
circuit in Fig. 10 does not offer any reduction of the secondary-side duty-cycle
loss, or circulating energy.
A number of techniques that can achieve ZVS of all primary switches in the
entire input-voltage and load range virtually without a loss of duty cycle and
CB
VIN
Q 1
Q 2
Q 3
Q 4
TR
o V
DR1
DR2
NP
NS
NS
L F
CF
RL
DC1
DC2
LP
Fig. 10. FB ZVS-PWM converter with primary-side clamp diodes.32
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
410 M. M. Jovanovic
secondary-side ringing have been proposed in Refs. 33–36. Generally, these cir-
cuits achieve ZVS of all primary switches in the entire load and input-voltage range
by utilizing energy stored in inductive components of an auxiliary circuit. In the
approach described and analyzed in Refs. 33 and 34, the auxiliary circuit comprises
of a pair of inductors that connected between the mid-point of the bridge legs
and a mid-point of an input-voltage capacitive divider, whereas in the approach
described in Ref. 35, the energy stored in a magnetizing inductance of an auxil-
iary transformer is used to extend the ZVS range. While in the FB ZVS-PWM
converters described in Refs. 33–35, the energy available for ZVS increases as the
input-voltage increases, which is the desirable direction of change since more en-
ergy is required to achieve ZVS at higher input voltages, the stored energy in these
FB ZVS converters is independent of load. As a result, they cannot optimally re-
solve the trade-off between power-loss savings brought about by a full-load-range
ZVS and power losses of the auxiliary circuit. Ideally, the auxiliary circuit needs
to provide very little energy, if any, at full load since the full-load current stores
enough energy in converter’s inductive components to achieve a complete ZVS of
all switches. As the load current decreases, the auxiliary circuit needs to provide
progressively more ZVS energy, with the maximum energy required at no load. A
FB ZVS-PWM converter that features this kind of adaptive energy storage in the
auxiliary circuit has been introduced in Ref. 36.
This converter, shown in Fig. 11, employs a primary-side coupled inductor
connected between the bridge legs to achieve a wide-range ZVS. Generally, the
secondary-side of the circuit in Fig. 11 can be implemented with any type of the
full-wave rectifier such as the full-wave rectifier with a center-tapped secondary,
the full-wave rectifier with current doubler, or the full-bridge full-wave rectifier.
However, in high-power applications, the current-doubler rectifier shown in Fig. 11
is usually used since it shows the best performance.37,38 The control of the circuit
in Fig. 11 is the same as the control of any other constant-frequency FB ZVS con-
verter, i.e., any of the integrated phase-shift controllers available on the market can
be used.
In the circuit in Fig. 11, the ZVS of leading-leg switches Q3 and Q4 is achieved
primarily by the energy stored in the magnetizing inductance of coupled inductor
LC , whereas the ZVS of the lagging-leg switches Q1 and Q2 is achieved by the
sum of the energy stored in the magnetizing inductance of LC and output filter
inductance LF . By properly selecting the value of the magnetizing inductance of
the coupled inductor, all primary switches in the converter in Fig. 11 can achieve
ZVS even at no load. Because in the circuit in Fig. 11 the energy required to
create ZVS conditions at light loads does not need to be stored in the leakage
inductance, the transformer leakage inductance can be minimized. As a result,
the loss of the duty cycle on the secondary-side is minimized, which maximizes the
turns ratio of the transformer and, consequently, minimizes the conduction losses. In
addition, the minimized leakage inductance of the transformer significantly reduces
the secondary-side ringing caused by the resonance between the leakage inductance
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 411
TR
LC
CB1
CB2
VIN
Q 1
Q 2
Q 3
Q 4
NC NC
NP
C1
D1
C2
D2
C3
D3
C4
D4
I O
D1
D2
o V
L F1
CF RL
L F2
NS
Fig. 11. FB ZVS-PWM converter with coupled inductors.36
and junction capacitance of the rectifier, which greatly reduces the power dissipation
of a snubber circuit that is usually used to damp the residual high frequency ringing.
For a properly selected maximum turns ratio of the transformer, the conduction
loss of the converter in Fig. 11 can be reduced by as much as 30% compared to
the corresponding loss in the conventional circuit in Fig. 8. Furthermore, because
of the ZVS operation across the entire load and line range, the converter in Fig. 11
can efficiently operate even at significantly higher frequencies than the circuit in
Fig. 8.
Generally, bridge-type topologies are not used in multiple output power sup-
plies that are supposed to deliver high currents from low-voltage outputs. Instead,
single-ended topologies are employed because they are more suitable for implemen-
tations with synchronous rectifiers (SRs) and mag-amp post regulators.39,40 The
topology choices in these implementations are limited to the single-switch and two-
switch forward converters and their variations. Their power density is primarily
limited by a relatively low maximum frequency at which mag-amps can efficiently
work. Typically, mag-amp performance rapidly deteriorates at frequencies above
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
412 M. M. Jovanovic
the 100–150 kHz range. Therefore, to improve the power density, it is necessary to
resort to implementations without mag-amps. This can be done either by system
partitioning where the number of post-regulated outputs is reduced by the em-
ployment of multiple number of power stages (converters), or by the employment
of dc/dc post regulators, or by their combination. As an example, 12-V, 5-V, and
3.3-V outputs can be generated by a converter that has a 12-V output and employs
two dc/dc step-down converters to create the 5-V and 3.3-V outputs.
4. Techniques for Hold-Up-Time Energy-Storage Minimization
As the required power and power density of ac/dc datacom power supplies increase,
the size of the hold-up-time energy-storage (bulk) capacitors, which store the energy
that is used to deliver power to the load for a short time (typically 10–20 ms) after
a line dropout, becomes a significant factor in limiting the maximum attainable
power density. Therefore, the minimization of the hold-up-time capacitor size is
an indispensable design step in maximizing the power density of ac/dc datacom
power supplies. The minimization of the size of the hold-up-time capacitors can
be accomplished by advancements in capacitor technology that would increase the
volumetric efficiency (µF/in3) of the energy-storage capacitors and/or by circuit
optimization techniques that would minimize the amount of capacitance needed for
a given hold-up-time. In this paper, only the circuit techniques that can be used to
minimize the energy-storage capacitors are discussed.
To achieve a desired hold-up-time, the dc/dc converter output stage must be de-
signed to operate in a certain voltage range with minimum energy-storage-capacitor
voltage VBMIN lower than voltage VBH that corresponds to the line voltage at which
hold-up-time is defined, as illustrated in Fig. 12. With such a design of the dc/dc
output stage, the energy storage capacitor delivers power to the output after a line
dropout until the energy-storage-capacitor discharges to VBMIN .
Generally, for a given VBH and VBMIN , larger power POH and/or longer hold-
up-time TH requires a larger energy-storage-capacitorCB . As a result, in high-power
V
B
T0
HOLD-UP TIME
t
MOMENT OF INPUT-VOLTAGE
LINE DROP OUT
V BH
V BMIN
T H
T1
Fig. 12. Energy-storage-capacitor voltage waveform during hold-up-time.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 413
V BH
V BMIN
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
E CBH
E CB
0.75
0.36
0.19
Fig. 13. Plot of ratio of delivered energy and total stored energy of hold-up-time capacitor∆ECB/∆ECBH as function of VBMIN /VBH .
applications, the size of energy-storage-capacitor(s) very often limits the maximum
power density. To maximize power density, the size of energy-storage-capacitors
must be minimized. A limited size reduction of energy-storage-capacitor CB can be
achieved by extending the regulation range of the dc/dc converter output stage by
minimizing voltage VBMIN at which the dc/dc converter output stage drops out
of regulation. However, because of a strong trade-off between minimum regulation
voltage VBMIN and the conversion efficiency of the dc/dc converter output stage,
VBMIN is usually restricted to 80% to 90% of VBH . With such a selection of VBMIN ,
only a small part of the energy stored in CB is delivered during the hold-up-time
period.
As illustrated in Fig. 13, which shows the percentage of the total stored energy
delivered to the load during the hold-up-time as a function of selected VBMIN , only
19% of the stored energy is delivered to the load during the hold-up-time if VBMIN
is selected to be 0.9VBH . Similarly, if VBMIN = 0.8VBH , 36% of the stored energy
is delivered to the output, i.e., still the majority of the stored energy is not used to
supply the load during the hold-up-time.
To utilize the majority of the stored energy during the hold-up-time, VBMIN
must be selected well below 80% of VBH . For example, 75% of the stored energy
is delivered to the load for VBMIN = 0.5VBH . However, with VBMIN = 0.5VBH ,
the efficiency of the dc/dc converter and, therefore, the overall efficiency would be
severally penalized because the dc/dc converter output stage would be required to
operate with a much wider input voltage range. Namely, to regulate the output
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
414 M. M. Jovanovic
in a wider input-voltage range, a wider duty-cycle range is needed, which dictates
the selection of a smaller turns ratio of the transformer. Generally, a smaller turns
ratio increases the primary and, very often, secondary conduction losses, which
deteriorates the conversion efficiency.
So far, two techniques for the minimization of the hold-up-time capacitor size
without the efficiency degradation are proposed and demonstrated. The technique
described in Ref. 41 employs variable transformer-turns-ratio concept, whereas in
the technique introduced in Ref. 42 employs a hold-up-time extension circuit to
maximize the utilization of the stored energy. The implementation of the variable-
turns-ratio circuit is shown in Fig. 14. In this circuit the effective turns ratio of
the transformer can be changed by secondary-side switch Q. Switch Q is turned on
only during the hold-up-time period, otherwise switch Q is off. When switch Q is
off, windings N2 are open so that the turns ratio is maximal at n = N/N1, which
maximizes the conversion efficiency. During the hold-up-time mode of operation
LLK
CB
VIN
Q 1
Q 2
Q 4
C1
D1
C2
D2
Q 3
C3
D3
C4
D4
TR
N
D1
D2
o V
L F1 CF
N1
L F2 Q
D3
D4
N2
N2
Fig. 14. FB ZVS-PWM converter with variable turns-ratio transformer.41
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 415
when switch Q is on, windings N2 conducts. As a result, the turns ratio of the
transformer is reduced to n = N/(N1 + N2), which allows the converter to operate
down to a lower bulk capacitor voltage so that the utilization of the energy stored
in the bulk capacitors is optimized. A potential problem of the circuit in Fig. 14
is a possible out-of-spec disturbance of the output-voltage caused by an abrupt
change of the turns ratio of the transformer at the moment switch Q is turned on.
In addition, this concept is not suitable for applications in multiple-output power
supplies because it would require a separate switch and windings for each output.
The conceptual block diagram of the hold-up-time extension circuit is shown in
Fig. 15(a). Generally, the improved utilization of the stored hold-up-time energy
is obtained by providing two energy storage capacitors, CBO and CBAUX , and
by connecting auxiliary energy-storage-capacitor CBAUX to the input of the hold-
up-time extension circuit that has the output connected to the energy-storage-
capacitor CBO . To ensure proper operation, diodes DBO and DBAUX are used to
provide necessary decoupling of certain parts of the circuits during different phases
of operation. It should be noted that depending on the actual implementation of
the front-end rectifier circuit, as well as the hold-up-time circuit, diode DBAUX
may not be required.
Since the hold-up-time extension circuit is connected at the input of the dc/dc
output stage, its performance and complexity is unaffected by the dc/dc output
stage topology and number of outputs. In addition, since the hold-up-time extension
circuit acts outside the control loop of the dc/dc output stage, the dc/dc output(s)
are virtually immune to the disturbance that may be generated at the moment the
hold-up-time extension circuit is activated.
The hold-up-time extension circuit is designed so that its output is regulated
at voltage VBAUX that is lower than the normal-mode-operation voltage VBH and
slightly higher than the minimum regulation voltage of the dc/dc converter VBMIN .
As a result, the hold-up-time extension circuit is inactive during the normal mode of
operation, i.e., when the ac line is present at the input of the power supply. In fact,
during the normal mode of operation, capacitors CBO and CBAUX are connected
in parallel so that their voltages are equal if the voltage drop across rectifier DBO
is neglected.
When the ac line drops out, the output power is first supplied by the energy
stored in capacitors CBO and CBAUX . As a result, the capacitors start discharging
and their voltages starts decreasing, as shown in Fig. 15(b). During this phase of
operation, voltage VBO across capacitor CBO and voltage VBAUX across capacitor
CBAUX track each other since the capacitors are simultaneously discharging. In
fact, the operation of the circuit during this phase is identical to that of the circuit
without the hold-up-time extension circuit. When the voltage across the capaci-
tors reaches a level close to the minimum regulation voltage of the dc/dc converter
VBMIN , i.e., when VBO = VBAUX ≈ VBMIN , the hold-up-time extension circuit be-
comes activated. Once the hold-up-time extension circuit is activated, the capacitor
CBAUX connected to the input of the hold-up-time extension circuit continues to
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
416 M. M. Jovanovic
V BOvIN
C
V O1
FRONT-END
RECTIFIER BO
V
.
.
.
O2
V On
V BAUX
C
HOLD-UP TIME
EXTENSION CIRCUIT
BAUX
DBO
DBAUX
V H = V BAUXMIN
DC/DC
CONVERTER
OUTPUT
STAGE
(a)
V B
DC/DC CONVERTER
REGULATION RANGE
T0
ORIGINAL
HOLD-UP TIME
t
V BH
V BMIN
T H
T1
HOLD-UP TIME
EXTENSION
T H
HOLD-UP TIME
V BAUXMIN
V BAUX
V BO
V BAUX
V BO
V BAUX
MOMENT OF INPUT-VOLTAGE
LINE DROP OUT
(b)
Fig. 15. (a) Conceptual implementation of hold-up-time extension circuit: block diagram and(b) energy-storage-capacitor voltage VBO and auxiliary energy capacitor voltage VBAUX duringhold-up-time.
discharge and provide energy to the regulated output of the hold-up-time extension
circuit, which keeps the voltage across capacitor CBO constant at approximately
VBMIN , as shown in Fig. 15(b). Therefore, during this phase of operation, energy
stored in capacitor CBAUX is used to delivered the required output power during
the hold-up-time. The hold-up-time is terminated when the voltage across CBAUX
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 417
decreases to VBAUXMIN , which is the minimum regulation voltage of the hold-up-
time extension circuit.
As can be seen from Fig. 15(b), the extension of the hold-up-time brought about
by the hold-up-time extension circuit is dependent on the selection of VBAUXMIN .
A lower VBAUXMIN results in a longer extension of the hold-up-time because more
energy is taken from capacitor CBAUX . For a given VBAUXMIN , the extension of
the hold-up-time depends on the size of CBAUX and it increases as the capacitance
of CBAUX is increased.
To evaluate the effectiveness of the proposed hold-up-time extension method,
Fig. 16 shows plots of the ratio of the additional discharge energy due to the
hold-up-time extension circuit ∆ECBEC and the energy discharge without the
hold-up extension circuit ∆ECB as functions of VBAUXMIN/VBH and for differ-
ent CBAUX/(CBO + CBAUX ) ratios. The solid lines in Fig. 16 present plots for
VBMIN/VBH = 0.9, whereas the dashed lines show plots for VBMIN/VBH = 0.8,
i.e., for two regulation ranges of the dc/dc converter output stage. The plots are
generated by assuming that CBO +CBAUX = CB , i.e., that total energy storage ca-
pacitance in the implementations with and without the hold-up-time circuit is the
same. Therefore, different ratios of CBAUX/CB used as the parameter in the plots
V BH
V BAUXMIN
E CB
E CBEC
X = 0.5
X = 0.75
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
C B
C AUX
X =
X = 0.25
X =0.25
X = 0.75
X = 0.5
Fig. 16. Plot of ratio of additional discharged energy due to hold-up-time extension circuit∆ECBEC and discharge energy without hold-up-time extension circuit ∆ECB as functions of
CAUX/CB and VBAUXMIN /VBH . Solid line for VBAUXMIN /VBH = 0.9, dashed line forVBAUXMIN /VBH = 0.8.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
418 M. M. Jovanovic
in Fig. 16 mean different allocations of the total capacitance CB among auxiliary
capacitance CBAUX and capacitance CBO .
As can be seen from Fig. 16, for a given ratio VBAUXMIN/VBH , the amount
of extracted energy from the energy-storage capacitors is increased as the ratio
X = CBAUX/CB increases, i.e., as more of the total capacitance is allocated to
CBAUX . For example, for VBMIN/VBH = 0.9 (solid lines), VBAUXMIN/VBH = 0.5,
and X = 0.25, the hold-up-time extension circuit helps extract approximately 70%
more energy from the energy-store capacitors compared with the implementation
without the hold-up-time extension circuit that has the same amount of the energy-
storage capacitance. For larger values of ratio X = CAUX/CB , the effect of the
hold-up-time extension circuit is even more dramatic since it increases the amount
of the delivered energy from the storage capacitors by approximately 150% and
220% for X = 0.5 and X = 0.75, respectively.
The effectiveness of the hold-up-time extension circuit is also dependent on the
regulation range of the dc/dc converter, i.e., it is dependent on the VBMIN/VBH
ratio. As can be seen from Fig. 16, for VBMIN/VBH = 0.8 (dashed lines) and
VBAUXMIN /VBH = 0.5, the hold-up-time extension circuit helps extract approxi-
mately 25%, 55%, and 80% more energy for X = 0.25, X = 0.5, and X = 0.75,
respectively. Generally, for the maximum effectiveness of the hold-up-time circuit,
it is necessary to maximize auxiliary capacitance CBAUX since the energy stored in
this CBAUX is delivered to the output when the hold-up-time extension circuit is
activated. The energy stored in capacitor CBO is only delivered during the hold-up-
time period before the hold-up-time extension circuit is activated at approximately
VBMIN . Therefore, the design optimization of the hold-up-time extension circuit
requires that the available total energy-storage capacitance be allocated between
CBO and CBAUX so that CBAUX is maximized. In practice, this allocation is usu-
ally dictated by the minimum capacitance CBO that can handle the voltage and
current ripple produced at the output of the hold-up-time extension circuit.
Because when the hold-up-time extension circuit is activated the voltage across
CBAUX is lower than the voltage across CBO, as shown in Fig. 15(b), the hold-
up-time extension circuit can be implemented with any boost-like nonisolated or
isolated topology. Figure 17 shows the implementation of the hold-up-time extension
circuit with the boost topology.
5. EMI Filter Optimization Issues
In today’s power supplies, the size and power loss of the EMI filter is emerging
as one of the major limitations to further power-density and efficiency improve-
ments. The primary reason is the lack of practical design techniques for size and
loss optimization, although a number of useful papers addressing various EMI mea-
surement, simulation, and component-value selection issues have been published in
recent years.43–48 As a result, the cut-and-try method is still the prevalent design
optimization technique used today in the power supply industry.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 419
V BOv
INC
V O1
FRONT-END RECTIFIER
BO
V
.
.
.
O2
V On
V BAUX
CBAUX
DL
S
DC/DC
CONVERTER
OUTPUT
STAGE
HOLD-UP TIME
EXTENSION CIRCUIT
LAUX
SAUX
DAUX
D5
D6
D1
D3
D2
D4
Fig. 17. Hold-up-time extension circuit implementation with boost converter.
Generally, to minimize the size of an EMI filter, it is not only necessary to
minimize the number of the filter stages by careful design of the circuit layout and
selection of filter component values, but also it is necessary to develop fabrica-
tion/packaging techniques that would minimize the size of filter’s magnetic com-
ponents such as differential and common-mode filters. Whereas component-value
optimization issues are being slowly but surely addressed, size optimization issues
are still waiting to get their due attention.
Typically, the dominant loss in EMI filters for high-power power supplies is the
copper loss of the windings of the differential and common-mode chokes. Therefore,
one of the major tasks in the optimization of the EMI filter performance and size is
to maximize the amount of copper so that the copper loss is reduced and conversion
efficiency maximized. The second task is to find efficient thermal-management and
packaging methods to extract the residual heat out of the filter, which would allow
for a further filter-size reduction.
Finally, the size of the capacitive components of an EMI filter such as X and Y
capacitors, which may become a power-density limiting factor once the size of filter’s
magnetic components is optimized, must be also reduced. While some reduction of
the X and Y caps can be gained through filter optimization, a major size reduction
is expected from the advancements in the capacitor technology. Perhaps, the future
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
420 M. M. Jovanovic
generation X and Y caps for use in high-power density EMI filters will be of a
surface-mounted type.
6. Summary
Ever-increasing power density requirements of modern ac/dc server and telecom
power supplies pose a formidable challenge to power supply designers. To succeed
in meeting these high-power density requirements, it is necessary to optimize per-
formance and size of all major functional parts of the power supply. Specifically,
the utmost attention needs to be paid to the performance and size optimizations
of the PFC front-end converter, dc/dc output stage, and EMI filter, as well to the
size optimization of the hold-up-time energy-storage capacitors.
While circuit optimization is certainly an indispensable ingredient in increasing
power density, the major future power-density gains are expected from advance-
ments in thermal management and packaging technologies that will enable higher
component integration levels. In addition, advancements in active and passive com-
ponent technologies such as new materials and packaging solutions will play a major
role in a continuous push for ever-higher power densities.
With the recent introduction of the SiC (Silicon Carbide) technology, which
makes it feasible to operate existing converters at significantly higher switching
frequencies, the bottleneck for size reduction is shifting from semiconductor devices
to passive components. In fact, at high switching frequencies, increased core and
winding losses in magnetic components and the volume required for the hold-up-
time energy-storage capacitors are emerging as the major factors that limit the
maximum attainable power density.
References
1. B. Griffith, Server power building blocks, Handout of the 5th Annual Intel Technology
Symposium, Seattle, WA, 27–28 September 2001.2. J. T. Strydom, J. D. van Wyk and M. A. de Rooij, Integration of a 1-MHz converter
with active and passive stages, IEEE Appl. Power Electron. (APEC ) Conf. Proc.
(2001), pp. 1045–1050.3. X. Liu, S. Haque, J. Wang and G. Q. Lu, Packaging of integrated power electronics
modules using flip-chip technology, IEEE Appl. Power Electron. (APEC ) Conf. Proc.
(2000), pp. 290–296.4. S. Haque, X. Liu and G. Q. Lu, Thermal management of power electronics modules
via acoustic micrography imaging, IEEE Appl. Power Electron. (APEC ) Conf. Proc.
(2000), pp. 299–305.5. Y. Khersonsky, M. Robinson and D. Gutierrez, New fast recovery diode technol-
ogy cuts circuit losses, improves reliability, Power Conversion & Intelligent Motion
(PCIM ) Magazine (May 1992), pp. 16–25.6. G. Carli, Harmonic distortion reduction schemes for a new 100A-48V power supply,
IEEE Int. Telecommunication Energy Conf. (INTELEC ) Proc. (1992), pp. 524–531.7. M. M. Jovanovic, Z. Chen and P. Liao, Evaluation of active and passive snubber tech-
niques for applications in power-factor-correction boost converters, Sixth Int. Conf.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 421
Power Semiconductors and Their Applications (Electronica ’92) Proc. (1992), pp. 105–115.
8. K. M. Smith and K. M. Smedley, A comparison of active and passive soft switchingmethods for PWM converters, IEEE Power Electronics Specialists’ Conf. (PESC )Rec. (1998), pp. 94–100.
9. K. M. Smith and K. M. Smedley, Engineering design of lossless passive soft switchingmethods for PWM converters, IEEE Appl. Power Electron. Conf. (APEC ) Proc.
(1998), pp. 1055–1061.10. S. Ben-Yakov and G. Ivensky, Passive lossless snubbers for high frequency PWM con-
verters, IEEE Appl. Power Electron. Conf. (APEC ) Professional Education Seminar
Book, Seminar 12 (1999).11. C. J. Tseng and C. L. Chen, Passive lossless snubbers for dc/dc converters, IEEE
Appl. Power Electron. Conf. (APEC ) Proc. (1998), pp. 1049–1054.12. R. Streit and D. Tollik, High efficiency telecom rectifier using a novel soft-switched
boost-based input current shaper, IEEE Int. Telecommunication Energy Conf.
(INTELEC ) Proc. (1991), pp. 720–726.13. G. Hua, C. S. Leu and F. C. Lee, Novel zero-voltage-transition PWM converters,
IEEE Power Electronics Specialists’ Conf. (PESC ) Rec. (1992), pp. 55–61.14. D. C. Martins, F. J. M. de Seixas, J. A. Brilhante and I. Barbi, A family of dc-
to-dc PWM converters using a new ZVS commutation cell, IEEE Power Electronics
Specialists’ Conf. (PESC ) Rec. (1993), pp. 524–530.15. H. Matsuo, F. Kurokawa, T. Oshikata and Y. Yamawaki, Analysis of dynamic charac-
teristics for the partially resonant active filter with the DSP, Int. Telecommunication
Energy Conf. (INTELEC ) Proc. (2001), pp. 81–88.16. J. Bassett, New, zero voltage switching, high frequency boost converter topology for
power factor correction, IEEE Int. Telecommunication Energy Conf. (INTELEC )Proc. (1995), pp. 813–820.
17. M. M. Jovanovic, A technique for reducing rectifier reverse-recovery-related losses inhigh-voltage, high-power boost converters, IEEE Applied Power Electronics (APEC )Conf. Proc. (1997), pp. 1000–1007.
18. C. M. C. Duarte and I. Barbi, An improved family of ZVS-PWM active-clampingdc-to-dc converters, IEEE Power Electronics Specialists’ Conf. (PESC ) Rec. (1998),pp. 669–675.
19. M. M. Jovanovic and Y. Jang, A new soft-switched boost converter with isolated activesnubber, IEEE Applied Power Electronics (APEC ) Conf. Proc. (1998), pp. 1084–1090.
20. K. Wang, F. C. Lee, G. Hua and D. Borojevic, A comparative study of switching lossesof IGBTs under hard-switching, zero-voltage-switching, and zero-current-switching,IEEE Power Electronics Specialists’ Conf. (PESC ) Rec. (1994), pp. 1196–1204.
21. G. Hua, X. Yang, Y. Jiang and F. C. Lee, Novel zero-current-transition PWM con-verters, IEEE Power Electronics Specialists’ Conf. (PESC ) Rec. (1993), pp. 538–544.
22. K. Wang, G. Hua and F. C. Lee, Analysis, design and experimental results of ZCS-PWM boost converters, Int. Power Electronics Conf. Proc. (1995), pp. 1202–1207.
23. Y. Jang and M. M. Jovanovic, A new, soft-switched, high-power-factor boost converterwith IGBTs, IEEE Int. Telecommunication Energy Conf. (INTELEC ) Proc., Paper8–3 (1999).
24. H. Kapels, R. Rupp, L. Lorenz and I. Zverev, SiC Schottky diodes: A milestone inhard switching, Power Conversion and Intelligent Motion Conf. (PCIM ) Proc. (2001),pp. 95–100.
25. J. Petzoldt, T. Reimann, M. Scherf and I. Zverev, Power losses of an ultra-fastcoolmos/SiC-diode device-set in PFC-applications: Simulation and measurement,Power Conversion and Intelligent Motion Conf. (PCIM ) Proc. (2001), pp. 115–120.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
422 M. M. Jovanovic
26. O. D. Petterson and D. M. Divan, Pseudo-resonant full bridge dc/dc converter, IEEE
Power Electronics Specialists’ Conf. (PESC ) Rec. (1987), pp. 424–430.27. J. Sabate, V. Vlatkovic, R. B. Ridley, F. C. Lee and B. H. Cho, Design considerations
for high-voltage high-power full-bridge zero-voltage-switched PWM converter, IEEE
Applied Power Electronics Conf. (APEC ) Proc. (1990), pp. 275–284.28. W. Chen, F. C. Lee, M. M. Jovanovic and J. A. Sabate, A comparative study of
a class of full-bridge zero-voltage-switched PWM converters, IEEE Applied Power
Electronics (APEC ) Conf. Proc. (1995), pp. 893–899.29. G. Hua, F. C. Lee and M. M. Jovanovic, An improved full-bridge zero-voltage-switched
PWM converter using a saturable inductor, IEEE Power Electronics Specialists’ Conf.
Rec. (1991), pp. 189–194.30. L. J. Hitchcock, M. M. Walters and R. A. Wunderlich, Resonant-transition dc-to-dc
converter, U.S. Patent 5,132,889, 21 July 1992.31. R. Watson and F. C. Lee, Analysis, design, and experimental results of a 1-kW
FB-ZVS-PWM converter employing mag-amp secondary-side control, IEEE Applied
Power Electronics (APEC ) Conf. Proc. (1994), pp. 166–172.32. R. Redl, N. O. Sokal and L. Balogh, A novel soft-switched full-bridge converter:
Analysis, design considerations and experimental results at 1.5 kW, 100 kHz, IEEE
Power Electronics Specialists’ Conf. Rec. (1990), pp. 162–172.33. M. Nakaoka, S. Nagai, Y. J. Kim, Y. Ogino and Y. Murakami, The state-of-the-art
phase-shifted ZVS-PWM series and parallel resonant dc–dc power converters usinginternal parasitic circuit components and new digital control, IEEE Power Electronics
Specialists’ Conf. (PESC ) Rec. (1992), pp. 62–70.34. P. K. Jain, W. Kang, H. Soin and Y. Xi, Analysis and design consideration of a load
and line independent zero voltage switching full bridge dc/dc converter topology,IEEE Trans. Power Electron. 17 (2002) 649–657.
35. R. Ayyanar and N. Mohan, Novel soft-switching dc–dc converter with full ZVS-rangeand reduced filter requirement — Part I: Regulated-output applications, IEEE Trans.
Power Electron. 16 (2001) 184–192.36. Y. Jang and M. M. Jovanovic, A new ZVS-PWM full-bridge converter, IEEE Int.
Telecommunication Energy Conf. (INTELEC ) Proc. (2002), pp. 232–239.37. N. Kutkut, A full-bridge soft-switched telecom power supply with current doubler rec-
tifier, IEEE Int. Telecommunication Energy Conf. (INTELEC ) Proc. (1997), pp. 344–351.
38. A. Pietkiewicz and D. Tolik, Coupled-inductor current-doubler topology in phase-shifted full-bridge dc–dc converter, IEEE Int. Telecommunication Energy Conf.
(INTELEC ) Proc. (1998), pp. 41–48.39. M. M. Jovanovic, M. Zhang and F. C. Lee, Evaluation of synchronous-rectification
efficiency improvement limits in forward converters, IEEE Trans. Power Electron. 42
(1995) 387–395.40. C. L. Jamerson, Post-regulation techniques for 100 kHz to 300 kHz multiple-output
PWM supplies, High-Frequency Power Conversion Conf. (HFPC ) Proc. (1989),pp. 260–273.
41. B. Yang, P. Xu and F. C. Lee, Range winding for wide input range front-end dc/dcconverter, IEEE Applied Power Electronics (APEC ) Conf. Proc. (2001), pp. 476–479.
42. Y. Jang and M. M. Jovanovic, Hold-up time extension circuits, U.S. Patent 6,504,497B2, 7 January 2003.
43. T. Guo, D. Chen and F. C. Lee, Separation of common-mode and differential-modeconducted EMI noise, High-Frequency Power Conversion Conf. Proc. (1994), pp. 128–134.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.
December 10, 2004 10:0 WSPC/123-JCSC 00165
Survey of State-of-the-Art Technologies for Computer–Network Server 423
44. F. Y. Shih, D. Y. Chen, Y. P. Wu and Y. T. Chen, A procedure for designing EMIfilters for ac line applications, IEEE Trans. Power Electron. 11 (1996) 170–181.
45. V. Vlatkovic, D. Borojevic and F. C. Lee, Input filter design for power factor correctioncircuits, IEEE Trans. Power Electron. 11 (1996) 190–205.
46. W. Zhang, M. T. Zhang, F. C. Lee, J. Roudet and E. Clavel, Conducted EMI analysisof a boost PFC circuit, IEEE Applied Power Electronics (APEC ) Conf. Proc. (1997),pp. 223–229.
47. Q. Chen, Electromagnetic interference (EMI) design considerations for a high-powerac/dc converter, IEEE Power Electronics Specialists’ Conf. (PESC ) Rec. (1998),1159–1164.
48. F. F. Li, EMC of power suppliers for telecommunications by design — A compu-tational approach, IEEE Int. Telecommunication Energy Conf. (INTELEC ) Proc.
(2001), pp. 636–640.
J C
IRC
UIT
SY
ST C
OM
P 20
04.1
3:39
9-42
3. D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by 8
.28.
25.3
on
02/1
5/13
. For
per
sona
l use
onl
y.