High Efficiency, High Density, PolyPhase Converters for High Current Applications

Application Note 77
September 1999
High Efficiency, High Density, PolyPhase Converters
for High Current Applications
Wei Chen
INTRODUCTION
As logic systems get larger and more complex, their
supply current requirements continue to rise. Systems
requiring 100A are fairly common. A high current power
supply to meet such requirements usually requires paralleling several power regulators to alleviate the thermal
stress on the individual power components. A power
supply designer is left with the choice of how to drive these
paralleled regulators: brute-force single-phase or smart
PolyPhaseTM.
A PolyPhase converter interleaves the clock signals of the
paralleled power stages, reducing input and output ripple
current without increasing the switching frequency. The
decreased power loss from the ESR of the input capacitor
and the low switching losses associated with MOSFETs at
relatively low switching frequencies help achieve high
power conversion efficiency. The size and cost of the input
capacitors are also greatly reduced as a result of input
ripple current cancellation. Since output ripple current
cancellation also occurs, lower value inductors can be
used. This results in improved dynamic response to load
transients. A combination of lower current rating and
decreased inductance also allows the use of smallersized, low profile, surface mounted inductors. For
multioutput applications, PolyPhase converters may also
provide the benefit of smaller input capacitors.
Previously, the implementation of multiphase designs
was difficult and expensive because of complex timing and
current-sharing requirements. The newly developed
LTC1629 solves these problems for high current, single
output designs, while the LTC1628 addresses dual-output
applications. Both ICs are dual, current mode, PolyPhase
controllers that can drive two synchronous buck stages
simultaneously. The features of the LTC1629 include a
unity-gain differential amplifier for true remote sensing,
low impedance gate drives, current-sharing, overvoltage
protection, optional overcurrent latch-off and foldback
current limit. Additionally, the LTC1629 can be configured
for 2-, 3-, 4-, 6- and 12-phase operation with a simple
phase selection signal (high, low or open). Optimizing the
number of phases can help achieve the smallest and the
most cost-effective power supply design.
This application note analyzes the performance of
PolyPhase converters and provides guidelines for selecting the phase number and designing a PolyPhase converter using the LTC1629. The following questions will be
answered as the discussion goes on:
• How much do I gain by using a PolyPhase architecture?
• How many phases do I need for my application?
• How do I design a PolyPhase converter?
HOW DO POLYPHASE TECHNIQUES EFFECT CIRCUIT
PERFORMANCE?
In general, PolyPhase operation improves the large signal
performance of a switched mode power converter, by
such means as reducing ripple current and ripple voltage.
A synchronous buck converter is used as an example in
this application note to analyze the effects of PolyPhase
techniques on circuit performance.
High current outputs usually require paralleling several
regulators. The single-regulator approach is not feasible
because of the unacceptable thermal stress on the individual power components. Paralleled regulators are synchronized to have the same switching frequency to eliminate
beat frequency noise at both the input and output terminals. Based on the phase relationship between the paral, LTC and LT are registered trademarks of Linear Technology Corporation.
PolyPhase is a trademark of Linear Technology Corporation.
AN77-1
Application Note 77
leled regulators, these converters can be divided into two
types: single-phase and PolyPhase. To balance the thermal stress in each component, paralleled regulators must
also share the load current.
In this application note, the number of channels refers to
the number of the paralleled regulators in one supply. The
following symbols are defined to facilitate reference:
• VO: DC output voltage
• IO: DC output current
• VIN: DC input voltage
• T: switching period
• mc: number of paralleled channels
• m: number of phases. The possible phase numbers are
usually determined by the channel number, mc. For
example, if mc = 6, the possible phase numbers are
m␣ =␣ 1, 2, 3, 6.
• CO: output capacitor
• ESR: equivalent series resistance of CO
• Lf: output inductor
Current-Sharing
The current-sharing can be easily achieved by implementing peak current mode control. In a current mode control
regulator, the load current is proportional to the error
voltage in the voltage feedback loop. If the paralleled
regulators see the same error voltage, they will source
equal currents. A 2-channel circuit is used as the example
to explain this current-sharing mechanism.
As shown in Figure 1, peak current mode control requires
that the high side switch turn off when the peak inductor
current (IL1, IL2) intersects the error voltage, VER, resulting in the same peak inductor currents. If the inductors are
identical, the peak-to-peak ripple currents of the inductors
will be the same. The DC currents of two inductors, which
are the peak current less half of the peak-to-peak ripple
current, will be equivalent. Two modules therefore share
the load current equally. The same current-sharing mechanism can be extended to any number of channels in
parallel. This current-sharing scheme will prevent an
individual module from suffering excessive current stress
in steady state operation and during line/load transient
conditions. Note that the sharing mechanism is open loop,
so no oscillations will occur due to current-sharing.
• D: duty cycle, approximated by VO/VIN in buck circuits
VGS(MAIN SWITCH OF MODULE 1)
IIN
O
IIN1
DT
T
VGS(MAIN SWITCH OF MODULE 2)
IN+
MODULE #1
IL1
CIN
+
L1
OUT+
+
IC
VER
CO
IN–
OUT–
IIN2
IO/2
IL1
MODULE #2
IL2
L2
IC
∆IO
AN77 F01a
(a)
AN77 F01b
(b)
Figure 1. 2-Channel Converter: (a) Schematic and (b) Typical Waveforms
AN77-2
IL2
Application Note 77
Output Ripple Current Cancellation and Reduced
Output Ripple Voltage
The phase relationship of Figure 1(b) shows how ripple
current cancellation at the output works. Because of the
180 degree phase difference between the two converters,
the two inductor ripple currents in the two-phase converter tend to cancel each other, resulting in a smaller
ripple current flowing into the output capacitor. The frequency of the output ripple current is doubled as well. All
of these factors contribute to a smaller output capacitor
for the same ripple voltage requirement.
Figure 2 shows the measured waveforms of the inductor
currents and output ripple currents in a 2-channel converter. The output ripple cancellation reduces the output
ripple current from 14AP-P (single-phase) to 6AP-P (dualphase). The ripple frequency in the dual-phase circuit is
twice the switching frequency.
To quantify the output ripple current amplitude in an mphase circuit, a closed-form expression was developed.
The derivation starts with the 2-phase circuit shown in
Figure 1. During the interval [from DT, T] when the high
side switch in module 1 is off and the high side switch of
module 2 is on, the inductor current in module 1 decreases
IL1
5A/DIV
and the inductor current in module 2 increases. The net
ripple current flowing into the output capacitors is smaller.
The output ripple current for the 2-phase circuit is derived
as:
∆IO =
2VO(1 − D)T 1 − 2D
Lf
1 − 2D + 1
(1)
See Appendix A for the detailed derivation procedure. By
extending the same derivation procedure to an m-phase
configuration, the output ripple current for an m-phase
circuit is obtained.
Output ripple current peak-to-peak amplitude in m-phase
circuit:
( )

VOT 1 − D

, m=1

Lf

m i

∆IO = 
∏ −D
i=1 m
mc • VOT •
, m = 2, 3,... (2)
 Lf
m−1 i

1

∏  −D + 
m

i=1  m
IL1
5A/DIV
IL2
5A/DIV
IL2
5A/DIV
IC
5A/DIV
IC
5A/DIV
AN77 F02a
(a) Single-Phase
AN77 F02b
(b) Dual-Phase
Figure 2. Output Ripple Current Waveforms In a 2-Channel Circuit. IL1 and IL2 Are the Inductor Currents In Two Channels
and IC Is the Net Ripple Current Flowing Into Output Capacitor. Test Conditions: VIN = 12V, VO = 2V, IO = 20A
AN77-3
Application Note 77
The output ripple voltage is estimated to be:
∆VO,PP <
∆IOT
+ ∆IO • ESR
8mCO
be rewritten as:
(3)
The first term in equation (3) represents the ripple voltage
on the pure capacitive components of CO and the second
term represents the ripple voltage generated on the ESR of
CO. Intuitively, a higher phase number helps reduce the
ripple component in the first term, and therefore, the
overall ripple voltage amplitude at the output. Another
interesting fact that can be observed is that the output
ripple current and voltage will reach zero if the duty cycle
is equal to one of the following critical points:
Dcrit =
i
,
m
i = 12
, ,...,m − 1
(4)
In a buck converter, the duty cycle is the ratio of the output
voltage and input voltage. By interpreting equation (4) in
terms of VIN and VO, the zero output ripple conditions can
VO
i
= ,
VIN m
, ,...,m − 1
i = 12
The plots in Figure 3 demonstrate the influence of phase
number and duty cycle on the output ripple current. In this
plot, the output ripple current is normalized against the
inductor ripple current at zero duty cycle (DIr = VOT/Lf).
The following assumptions are made: the number of
channels equals the phase number, the output voltage is
fixed and the power conversion efficiency is assumed to be
100%. This plot can be used to estimate the output ripple
current without tedious calculations.
The output ripple current approaches zero when the duty
cycle is near the critical points for the selected phase
number. For a buck circuit, the duty cycle is approximately
the ratio of VO/VIN. Therefore, if the input and output
voltages are relatively fixed, there exists an optimum
phase number to minimize the output ripple voltage.
1.00
0.95
1-PHASE
2-PHASE
3-PHASE
4-PHASE
6-PHASE
0.90
0.85
0.80
PEAK-TO-PEAK OUTPUT RIPPLE CURRENT
DIr
0.75
0.70
0.65
0.60
0.55
0.50
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
DUTY CYCLE (VO/VIN)
AN77 F03
Figure 3. Normalized Output Ripple Current vs Duty Cycle, DIr =
AN77-4
(5)
VOT
Lf
Application Note 77
Assuming that the maximum available phase number is
six and the efficiency is 100%, the optimum phase numbers for some common input and output voltages are
shown in Table 1.
Table 1. Optimum Phase Number for Minimizing the Ripple
Currents (Assuming That the Maximum Phase Number Is 6 and
the Efficiency Is 100%)
VO = 1.2V
VO = 1.5V
VO = 2.0V
VO = 2.5V
VIN = 5V
4
6
5
2, 4, 61
VIN = 12V
6
6
6
5
1 6 is the optimum phase number for minimum input ripple current.
For high step-down ratio or low duty cycle applications
(for example, VIN = 12V, VO = 1.2V, D = 0.1), a high phase
number helps reduce the maximum ripple current. For
wide duty cycle range applications, high phase number
tends to, but does not necessarily, yield lower output
ripple current. The optimum phase number needs to be
evaluated over the complete operating duty cycle range.
The reduction in the ripple current by increasing phase
number is not significant above four phases in most duty
cycle ranges.
Figure 4 shows the measured output ripple current near
the critical duty cycle point, which is Dcrit = 0.5 for a
2-phase circuit. The test conditions were VIN = 5V, VO = 2V,
IO = 20A, fs = 250kHz. Because of the voltage drop across
the MOSFET switches, the operating duty cycle was very
close to 50%. The dual-phase technique was able to
reduce the output ripple current considerably, from 10AP-P
(in the single-phase circuit) to 2.5AP-P. As a result, the
output ripple voltage near the critical duty cycle point is
negligible, as shown in Figure 5.
IL1
5A/DIV
IL1
5A/DIV
IL2
5A/DIV
IL2
5A/DIV
IC
5A/DIV
IC
5A/DIV
AN77 F04a
AN77 F04b
(a) Single-Phase
(b) Dual-Phase
Figure 4. Experimental Waveforms of Output Ripple Current Near Critical Duty Cycle Point (VIN = 5V, VO = 2V, IO = 20A, fs = 250kHz):
Top Trace, Inductor 1 Current; Middle Trace, Inductor 2 Current; Bottom Trace, Output Ripple Current
VOAC
10mV/DIV
VOAC
10mV/DIV
VSW1
10mV/DIV
VSW1
10mV/DIV
VSW2
10mV/DIV
VSW2
10mV/DIV
(a) Single-Phase
AN77 F05a
AN77 F05b
(b) Dual-Phase
Figure 5 Measured Output Ripple Voltage (Top Trace) near Critical Duty Cycle Point (VIN = 5V, VO = 2V,
IO = 20A, fs = 250kHz, VSW1 and VSW2 are the switch node voltages across the bottom FETs)
AN77-5
Application Note 77
Improved Load Transient Response
The influences of PolyPhase techniques on the load transient performance are numerous. First, the reduced output ripple voltage allows more room for voltage variations
during the load transient because the ripple voltage will
consume a smaller portion of the total error budget. With
the same number of capacitors on the output terminals of
the power supply, the sum of the overshoot and undershoot can be reduced dramatically. Second, the reduced
ripple current allows the use of lower value inductors. This
speeds up the output current slew rate of the power
supply. Consequently, PolyPhase helps improve the load
transient performance of the power supply. Figure 6
shows the output voltages during a load transient. It is
noted that the two circuits have the same electrical design.
The dual-phase technique reduces the voltage variation
from 69mVP-P to 58mVP-P, a 16% reduction with no
changes in component values. The inductor values could
be reduced while still achieving lower output ripple voltage
than the single-phase design, and further improvement in
the peak-to-peak voltage variations for the load transient
response could be realized.
currents. In a PolyPhase circuit, however, the paralleled
buck stages switch at different times and the pulsating
current flowing through the input capacitor is reduced
dramatically.
Figure 7 shows the measured input ripple current in a
2-channel converter. The PolyPhase converter reduces
the peak amplitude of the input ripple current by half and
doubles the ripple frequency. The reduced ripple current
amplitude results in a much smaller RMS current in the
input capacitor. Because the power loss on the ESR of the
input capacitor is proportional to the square of the RMS
current, the loss reduction can be significant. The size of
the input capacitor is reduced and the life of the capacitor
will likely be improved. The increased ripple frequency and
the reduced ripple amplitude also facilitate EMI filtering.
In order to quantitatively evaluate the input ripple current
in an m-phase circuit, a close-form expression is derived
by using some mathematical manipulations on the input
ripple current waveforms.
Input ripple current RMS value:
k  k + 1  2
mc2  Vo (1 − D)T 

− D Io +
D −  

 •


m  m
Lf
12mD2 

2
Input Ripple Current Cancellation
The input current of a buck converter is discontinuous.
With the input supply mainly sourcing DC current, the
input capacitor supplies a pulsating current to the buck
converter. In a single-phase circuit, the high side switches
of the paralleled buck modules turn on simultaneously.
The input capacitor needs to provide the sum of the pulsed
VOAC
20mV/DIV
Iirms =
3
3

k
k +1  
2
(k + 1)  D −  + k2 
− D 

 m
 
m


(6)
where k = FLOOR(m•D),
m = 1, 2, …
VOAC
20mV/DIV
AN77 F06a
(a) Single-Phase
AN77 F06b
(b) Dual-Phase
Figure 6. Measured Output Voltage During Load Transients (VIN = 12V, VO = 2V, fs = 250kHz.
Load steps: 5A to 20A and 20A to 5A, 50µs rise and fall times. Time scale: 500µs/DIV)
AN77-6
Application Note 77
The variable k is determined by the phase number (m) and
the duty cycle (D). For example, in a five-phase converter,
at 45% duty cycle, k = FLOOR(5 • 0.45) = 2. The FLOOR(x)
function provides a greatest integer that is smaller than or
equal to x.
As indicated in equation (6), the input ripple current of a
PolyPhase converter consists of two major factors: the DC
load current (first term) and the inductor ripple current
(second term). Since the inductor ripple current is almost
unaffected by the load conditions, the maximum RMS
input ripple current is reached at full load.
Usually, the size of the input capacitor is determined by the
power dissipation on its ESR. The full load condition
contributes to a maximum RMS input ripple current, and
therefore, determines the size of the input capacitor.
Figure 8 plots the RMS input ripple current against the
duty cycle for different phase configurations. In this plot,
the RMS input ripple current is normalized against the DC
load current. The output voltage is assumed fixed at 5V
and the input voltage is varied, resulting in a duty cycle
range from 0.1 to 0.9. Several facts can be observed from
the curves.
IIN1
IIN1
10A/DIV
10A/DIV
IIN2
IIN2
10A/DIV
10A/DIV
TOTAL IIN
TOTAL IIN
10A/DIV
10A/DIV
AN77 F07a
(a) Single-Phase
AN77 F07b
(b) Dual-Phase
Figure 7 Measured Input Ripple Current: Iin1 and Iin2 are the ripple currents into the paralleled modules.
Total Iin is the net ripple current into the input capacitor. (VIN = 12V, VO = 2V, IO = 20A, fs = 250kHz)
0.60
0.55
0.50
1-PHASE
2-PHASE
3-PHASE
4-PHASE
6-PHASE
RMS INPUT RIPPLE CURRNET
DC LOAD CURRENT
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
DUTY FACTOR (VO/VIN)
AN77 F08
Figure 8. Normalized RMS Input Ripple Current
AN77-7
Application Note 77
When the duty cycles are close to the critical duty cycle
points (determined in equation (4)), the first term in
equation (6) is zero. The RMS input ripple current reaches
the local minimum values. These values are not zero due
to the output inductor ripple current. Consequently, there
exists an optimum phase number to achieve the minimum
RMS input ripple current for a fixed input and output
application. For some common input and output voltages,
the optimum phase numbers for minimizing the input
ripple current are shown in Table 1. Note that these are the
same as the values for the minimum output ripple voltage.
For a wide duty cycle range application, higher phase
number helps reduce the maximum input ripple current.
But the reduction in the input ripple current by increasing
phase number may not be significant at higher phase
numbers in certain duty cycle ranges. The optimum phase
number needs to be evaluated over the complete operating
duty cycle range.
Figure 9 shows the experimental waveform of the input
ripple currents in a 2-channel circuit. The circuit operated
at close to 50% duty cycle, which is the critical duty cycle
point for the two-phase circuit. Compared to the singlephase technique, the PolyPhase technique reduced the
ripple current in the input capacitor dramatically.
DESIGN CONSIDERATIONS
Similar to the design of conventional paralleled regulators,
the design of a PolyPhase converter involves the choice of
the number of paralleled channels and the selection of the
power components (MOSFETs, inductors, capacitors, etc.).
Usually, the number of phases is set to be equal to the
number of channels. However, the number of channels
and the number of phases may be different. The number
of channels is usually determined by the total load current
and the acceptable current stress in each channel. For
example, if the required load current is 60A and the
maximum current stress per channel is 15A, 4 channels
need to be paralleled. The number of phases, on the other
hand, can be selected to minimize the input and output
filter capacitors. Note that each phase must have an equal
number of channels. In this example, a 4-channel configuration, one, two or four phases may be used.
Selection of phase number
As discussed in the previous section, the selection of a
different number of phases will greatly affect the input and
output ripple current.
For a narrow input and output range, the duty cycle range
is relatively narrow. The optimum phase number should
be chosen such that the circuit operates at or near one of
the critical duty cycle points (determined by equation 4).
For some practical input voltages and output voltages, the
optimum phase numbers for the minimum input ripple
currents and the lowest output ripple voltage are listed in
Table 1. For wide input or output voltage range, the phase
number should be chosen such that the worst-case RMS
input ripple current and the worst-case output ripple
voltage are minimized for the complete operating duty
cycle range.
IIN1
10A/DIV
IIN1
10A/DIV
IIN2
10A/DIV
IIN2
10A/DIV
TOTAL IIN
10A/DIV
TOTAL IIN
10A/DIV
(a) Single-Phase
AN77 F09a
(b) Dual-Phase
Figure 9. Input Current Near Critical Duty Cycle In a 2-Channel (VIN = 5V, VO = 2V, IO = 20A, fs = 250kHz)
AN77-8
AN77 F09b
Application Note 77
PolyPhase Converters using the LTC1629
Table 2. Phase Function Table for LTC1629
The LTC1629 integrates proprietary phase-locked-loopbased phasing circuitry. Each IC can be synchronized to an
external signal at the PLLIN pin and produce a CLKOUT
signal to synchronize other ICs. Table 2 shows the phase
function table of the LTC1629. By applying the command
signal (INTVCC, open or SGND) to the PHASMD pin and
connecting the CLKOUT pin of one IC to the PLLIN pin of
the next one, different numbers of phases can be achieved.
Figure 11 shows 2-phase, 3-phase, 4-phase, 6-phase and
12-phase configurations using the LTC1629.
PHASMD
0V
OPEN
INTVCC
PLLIN
0°
0°
0°
CONTROLLER 1
0°
0°
0°
CONTROLLER 2
180°
180°
240°
CLKOUT
60°
90°
120°
is provided by a 6-phase power supply, the two power
supplies can be interleaved by using a 12-phase configuration. As shown in Figure 12, U1, U2 and U3 are used to
produce the 3.3V output, and U4, U5, and U6 are used for
the 5V output. The resulting input ripple current frequency
is twelve times the switching frequency and the ripple
current amplitude is reduced.
A higher number of phases is usually needed for very high
output current or multioutput applications. For example,
in a 2-output system, 3.3V/90A and 5V/60A, if each output
U1
0
U1
U2
PHASMD
TG1
0° INTVCC
PHASMD
TG1
0°
PLLIN
TG2
180°
PLLIN
TG2
240°
CLKOUT
(a) 2-PHASE
28 CLKOUT
SENSE1 +
2
27 TG1
SENSE1 –
3
26 SW1
EAIN
4
25 BOOST1
PLLFLTR
5
24 VIN
PLLIN
6
23 BG1
PHASMD
7
22 EXTVCC
ITH
8
21 INTVCC
SGND
9
20 PGND
VDIFFOUT 10
18 BOOST2
VOS + 12
17 SW2
SENSE2 – 13
16 TG2
SENSE2 + 14
15 AMPMD
TG1
0°
PLLIN
TG2
180°
0
CLKOUT
PHASMD
TG1
90°
PLLIN
TG2
270°
CLKOUT
90°
U1
0
U2
PHASMD
TG1
0°
PLLIN
TG2
180°
0
U3
PHASMD
TG1
60°
PLLIN
TG2
240°
CLKOUT
0
PHASMD
TG1
120°
PLLIN
TG2
300°
CLKOUT
60°
CLKOUT
120°
(d) 6-PHASE
U1
0
U2
PHASMD
TG1
0°
PLLIN
TG2
180°
0
U3
PHASMD
TG1
60° OPEN
PHASMD
TG1
120°
PLLIN
TG2
240°
PLLIN
TG2
300°
210°
CLKOUT
CLKOUT
60°
0
CLKOUT
120°
U4
Figure 10. Pinouts of LTC1629
120°
(c) 4-PHASE
19 BG2
VOS – 11
TG2
U2
PHASMD
CLKOUT
1
TG1
PLLIN
(b) 3-PHASE
TOP VIEW
RUN/SS
PHASMD
120
U1
OPEN
0
U5
PHASMD
TG1
210°
PLLIN
TG2
30°
CLKOUT
0
U6
PHASMD
TG1
270°
PLLIN
TG2
90°
CLKOUT
0
PHASMD
TG1
330°
PLLIN
TG2
150°
CLKOUT
330°
270°
(e) 12-PHASE
AN77 F11
Figure 11. Configuration of Different Phases Using the LTC1629
AN77-9
Application Note 77
across the individual modules’ inputs and outputs (such
as A1B1, A2B2, etc.). The traces (AA1, AA2, BB1, BB2,
etc.) between modules should be the shortest and widest
possible to balance the current stress in each capacitor.
The impedance of the traces highlighted in Figure 13
should be minimized. It is preferable that these traces be
large copper planes. It is also important that the sources
of bottom MOSFETs (B1, B2, etc.) be connected to the
input filter capacitors before joining the ground plane
(CD). Otherwise, the ground noise generated by the pulsating current through the trace inductance will be seen on
the output terminals as spikes.
The LTC1629 includes a unity gain differential amplifier,
enabling true remote sensing of the output voltage. This is
particularly useful for maintaining tight output voltage
regulation for high current applications. Each LTC1629
based regulator consists of two synchronous buck stages
and two or more power regulators can be paralleled
directly. The inherent peak current mode control permits
automatic current-sharing. When several LTC1629-based
regulators are in parallel, the LTC1629 of the master
regulator senses the output voltage (VO +, VO –) via its
on-chip differential amplifier and divides this voltage
(VDIFFOUT) down through the resistor divider to utilize the
built-in 0.8V reference for output voltage regulation. This
control voltage is then fed to the EAIN pins (error amplifier
input) of each LTC1629. Since the error amplifier inside
the LTC1629 is a gm transconductance amplifier, directly
paralleling the ITH pins (error amplifier outputs) and the
EAIN pins is allowed. The paralleled regulators now share
the same error voltage. Because the load current of a
current mode regulator is proportional to the error voltage, the paralleled regulators must source equal currents.
IN+
A
A2
A1
E
CIN
+
To take full advantage of the ripple cancellation of the
PolyPhase technique, the input capacitors and output
capacitors are ideally placed at the summation points of all
the input ripple currents and all the output ripple currents,
respectively. Figure 13 shows the layout for a 2-phase
converter. In practice, the filter capacitors may be placed
+
CO
D
U1
0
B
B1
AN77 F13
IN–
C
Figure 13. Layout Diagram of Power Stage for 2-Phase Converter
0°
U2
PHASMD
TG1
0°
PLLIN
TG2
180°
0
U3
PHASMD
TG1
60° OPEN
PHASMD
TG1
120°
PLLIN
TG2
240°
PLLIN
TG2
300°
210°
CLKOUT
CLKOUT
0
CLKOUT
120°
60°
U4
U5
PHASMD
TG1
210°
PLLIN
TG2
30°
CLKOUT
0
OUT–
B2
OUTPUT #1
U6
PHASMD
TG1
270°
PLLIN
TG2
90°
CLKOUT
0
PHASMD
TG1
330°
PLLIN
TG2
150°
CLKOUT
330°
270°
OUTPUT #2
AN77 F12
Figure 12. 2-Output System Using 12-Phase Configuration
AN77-10
OUT+
E1
F
Layout Considerations
E2
Application Note 77
DESIGN EXAMPLE: 100A POLYPHASE POWER
SUPPLY
The specifications for a high current PolyPhase power
supply are as follows:
• Input: 12V (±10%)
• Outputs: 3.3V at 90A nominal, 100A maximum
• Load regulation: <20mV from 0A to full load
• Switching Noise: peak-to-peak voltage <1% of DC
voltage
• Efficiency:
• >89% at VIN = 12V, VO = 3.3V, IO = 90A;
Design Details
To utilize off-the-shelf, surface mount inductors and to
avoid the use of very thick PCB copper traces, it is
desirable to limit the individual module current to about
16A. Six channels are needed for this application. The
design now becomes only a 15A regulator which gets
repeated six times.
MOSFETs
The selection of MOSFETs is determined by the current
requirement and switching frequency. Low RDS(ON)
MOSFETs usually drive down the conduction losses but
tend to introduce high switching related losses at high
switching frequencies because of the large gate charge
and parasitic capacitances. For a given current requirement and selected switching frequency, both RDS(ON) and
gate charge (Qg) should be evaluated to minimize the sum
of the conduction losses, driving losses and the switching
loss. For the specified application, a number of MOSFETs
can be good choices: Si4420 (Siliconix), FDS6670A
(Fairchild), FDS7760A (Fairchild) and IRF7811 or IRF7805
(International Rectifier). In this application, we need two
MOSFETs for each high side switch and three MOSFETs
for each low side switch. The power dissipation in the
MOSFETs includes the conduction loss, switching loss
and reverse recovery loss of the body diodes in the low
side MOSFETs. The gate driving loss is seen by the
controller IC. In this design, if Si4420s are used, each top
MOSFET dissipates about 0.5W and each low side MOSFET
consumes about 0.9W. Based on the thermal resistance
provided in the datasheet, 30°C/W junction-to-ambient,
the maximum junction temperature of the MOSFETs is
about 30°C above the ambient temperature. Refer to the
LTC1629 data sheet and the literature from the MOSFET
vendors for more information on estimating the power
loss of MOSFETs.
Inductors
The selection of inductors is driven by the load current
amplitude and the switching frequency. The LTC1629
senses the inductor current with a current sense resistor.
The inductor ripple current must be large enough to
produce a reasonable AC sense voltage on the small value
sense resistor required for a high current application. A
reasonable starting point is to choose an inductor such
that its ripple current amplitude is about 40% of the
maximum channel current. It is estimated that an inductor
value between 1.0µH and 1.6µH will be suitable for 3.3V
output at a frequency of 200kHz. A number of off-theshelf, surface mount inductors are available for the specified application: P1608 (Pulse), PE53691 (Pulse),
ETQP6F1R3L (Panasonic) and CEPH149-1R6MC
(Sumida). Any inductor of similar inductance value and
current capability should work correctly.
AN77-11
Application Note 77
The maximum available phase number is six and the
possible phase number options are 1, 2, 3 and 6. By using
equations (1~6), the ripple currents at the input and output
for different-phase configuration are estimated as in
Table␣ 3.
The output capacitors are KEMET (T510X477M006AS
470µF/6.3V), ultralow ESR (30mohm), surface mount
tantulum capacitors. The peak-to-peak ripple current is
estimated to be 2.1AP-P. It would be 57.1AP-P if a conventional single-phase approach were taken instead.
A six-phase configuration minimizes both the input capacitor size and output ripple voltage. Compared to the
single-phase technique, the six-phase technique reduces
the input ripple current by more than 81%, and the output
ripple by more than 96%. Therefore, the six-phase configuration was adopted for this design.
Test results
Capacitors
The selection of the input capacitors is driven by the RMS
value of the input ripple current. Large capacitor ripple
currents introduce high power losses due to the ESR of the
capacitor. The resulting internal heating tends to reduce
the capacitor life. Low ESR capacitors must be used. This
design uses Sanyo OS-CON capacitors (16SA150M 150µF/
16V) whose maximum allowable ripple current is rated at
about 3.26ARMS. The input RMS ripple current for the sixphase configuration is estimated to be about 8.5ARMS.
Therefore, a minimum of three OS-CON capacitors are
needed. If a conventional single-phase technique were
used instead, the input RMS ripple current would be about
46.8ARMS. A minimum of fifteen OS-CON capacitors would
then be needed. Therefore, the use of an LTC1629 based
PolyPhase design saves at least twelve (15 – 3 = 12)
OS-CON capacitors.
The complete schematic diagram is shown in Figure 14.
The power supply consists of six buck channels and three
LTC1629s. Figure 15 shows the measured waveforms of
gate voltages and switch-node voltages in a typical sixphase converter. The gate voltages and switch-node voltages of six buck stages are interleaved 60 degrees output
of phase. Since the inductor current is driven by the
difference between the switch-node voltage and the DC
output voltage, the ripple currents in the six inductors are
also 60 degrees out of phase. As a result, the amplitude of
the net ripple current flowing into the output capacitor
decreases substantially and the ripple frequency increases
to six times the switching frequency. The output switching
noise and the power loss from the ESR are greatly attenuated.
Figure 16 shows the output voltage, which was measured
at the output capacitor (C14 in the schematics). The output
ripple voltage is less than 10mVP-P at 90A output current
and the ripple frequency is six times the switching frequency.
Efficiency was measured under different loading conditions. Figure 17 shows the efficiency curve. For most of the
load range, the efficiency was about 90%. At 100A, the
efficiency was measured 89.4%.
Table 3. Input and Output Ripple Current for Different Phase Configurations
Channels
6
6
6
6
Phases
1
2
3
6
Input Ripple Current (ARMS)
46.8
25.7
15.2
8.5
Output Ripple Current (AP-P)
57.1
19.0
6.3
2.1
Assuming that the efficiency is 100%, the inductance is 1.3µH, frequency is 200kHz
AN77-12
Application Note 77
C2
1nF
C1
1µF
1
2
3
4
C7, 47pF
C8, 1nF
INTVCC
5
6
7
R3, 47k
8
9
10
11
C11
470pF
R4
8.06k
12
13
14
R5
25.5k
CLKOUT
RUN/SS
SENSE1 +
TG1
SENSE1 –
SW1
BOOST1
EAIN
VIN
PLLFLTR
PLLIN
BG1
PHASMD
EXTVCC
ITH
INTVCC
SGND
PGND
VDIFFOUT
VOS –
VOS
BG2
BOOST2
+
SW2
SENSE2 –
TG2
SENSE2 +
AMPMD
28
27
4
R1, 10Ω
CLK1
VIN +
L1
1.3µH
C5
0.47µF
26
25
+
Q1
2× Si4420
C3
2µF
+
C4
VIN –
R2
0.003Ω
24
4
C6, 1µF
23
22
2
INTVCC
21
20
3
C9
2.2µF
19
+
C10
10µF
D2
1 BAT54A
C13
470µF
×9
D1
MBRS340T3
Q2
3×
Si4420
+
VO +
C14
10µF
VO –
18
17
R7
51Ω
C12, 0.47µF
16
4
15
L2
1.3µH
Q3
2× Si4420
U1, LTC1629
4
R8
51Ω
R6
0.003Ω
D3
MBRS340T3
Q4
3× Si4420
C15
1nF
VOSENSE +
VOSENSE –
C16
1nF
1
2
3
C17, 47pF
4
5
CLK1
C20, 47pF
6
7
8
9
R15, 10k
10
C34
0.01µF
11
12
13
C35, 1nF
14
CLKOUT
RUN/SS
SENSE1 +
TG1
SENSE1 –
SW1
BOOST1
EAIN
VIN
PLLFLTR
PLLIN
BG1
PHASMD
EXTVCC
ITH
INTVCC
PGND
SGND
BG2
VDIFFOUT
VOS
–
BOOST2
VOS
+
SW2
SENSE2 –
TG2
SENSE2 +
AMPMD
28
27
4
R9, 10Ω
CLK2
Q5
2× Si4420
L3
1.3µH
C18
0.47µF
26
25
24
23
4
C19, 1µF
22
2
3
21
20
19
C21
2.2µF
+
C22
10µF
D5
1 BAT54A
Q6
3×
Si4420
18
17
C23, 0.47µF
16
4
15
L4
1.3µH
Q7
2× Si4420
4
2
3
C26, 47pF
4
5
CLK2
C29, 47pF
6
7
8
9
10
R16, 10k
C37, 1nF
C36
0.01µF
11
12
13
14
CLKOUT
RUN/SS
SENSE1
+
TG1
SENSE1 –
SW1
BOOST1
EAIN
VIN
PLLFLTR
PLLIN
BG1
PHASMD
EXTVCC
ITH
INTVCC
PGND
SGND
BG2
VDIFFOUT
–
BOOST2
VOS +
SW2
SENSE2 –
TG2
SENSE2 +
AMPMD
VOS
D6
MBRS340T3
28
4
R12, 10Ω
27
Q9
2× Si4420
L5
1.3µH
C27
0.47µF
26
25
R13
0.003Ω
24
23
4
C28, 1µF
22
2
3
21
20
19
C30
2.2µF
+
C31
10µF
D8
1 BAT54A
D7
MBRS340T3
Q10
3×
Si4420
18
17
16
15
C32, 0.47µF
4
Q11
2× Si4420
L6
1.3µH
U3, LTC1629
4
C33
1nF
R11
0.003Ω
Q8
3× Si4420
C24
1nF
1
C4: THREE SANYO OS-CON 16SA150M
C13: NINE KEMET T510E477M006AS
L1 – L6: PANASONIC ETQP6F1R3L
D4
MBRS340T3
U2, LTC1629
C25
1nF
R10
0.003Ω
R14
0.003Ω
D9
MBRS340T3
Q12
3× Si4420
AN77 F14
Figure 14. Schematic Diagram of 3.3V/100A Six-Phase Converter
AN77-13
Application Note 77
C1 FREQUENCY
196.96kHz
C1 FREQUENCY
206.72kHz
AN77 F15a
AN77 F15b
(a) Voltage Scale: 5V/DIV, Time Scale: 1µs/DIV
(b) Voltage Scale: 10V/DIV, Time Scale: 1µs/DIV
Figure 15. Typical Waveforms of 6-phase Converter: (a) Gate Voltages,
(b) Switch-Node Voltages (Drain-to-Source Voltage of the Synchronous Switch)
10V/DIV
10mV/DIV
AN77 F16
Figure 16. Output Ripple Voltage Waveforms (Time Scale: 1us/DIV): VIN = 12V, VO = 3.3V, IO = 90A
100
EFFICIENCY (%)
90
12VIN, 3.3VO
80
70
60
50
0
10 20 30 40 50 60 70 80 90 100
LOAD CURRENT (A)
AN77 F17
Figure 17. Measured Efficiency at Different Load
AN77-14
Application Note 77
SUMMARY
PolyPhase converters reduce the input and output ripple
currents by interleaving the clock signals of paralleled
power stages. With a proper choice of phase number, the
output ripple voltage and the input capacitor size can be
minimized without increasing the switching frequency.
Lower output ripple voltage and smaller output inductors
help improve the circuit’s dynamic performance during
load transients. The low switching losses and driving
losses of MOSFETs at relatively low switching frequency,
and the reduced power losses due to ESRs of the capacitors, help achieve high efficiency.
The LTC1629 dual, PolyPhase current mode controller is
able to achieve the benefits of the PolyPhase technique
without complicating the control circuit. The LTC1629
helps minimize the external component count and simplify
the complete power supply design by integrating two
PWM current mode controllers, true remote sensing,
selectable phasing control, current-sharing capability,
high current MOSFET drivers and protection features
(such as overvoltage protection, optional overcurrent
latch-off and foldback current limit) into one IC. The
resulting manufacturing simplicity helps improve power
supply reliability. The high current MOSFET drivers allow
the use of low RDS(ON) MOSFETs to minimize the conduction losses for high current applications. Lower current
ratings on the individual inductors and MOSFETs also
make it possible to use low profile, surface mount components. Therefore, an LTC1629-based PolyPhase high
current converter can achieve high efficiency, small size
and low profile simultaneously. The savings on the input
and output capacitors, inductors and heat sinks minimize
the overall cost and size of the complete power supply.
Information furnished by Linear Technology Corporation is believed to be accurate and reliable.
However, no responsibility is assumed for its use. Linear Technology Corporation makes no representation that the interconnection of its circuits as described herein will not infringe on existing patent rights.
AN77-15
Application Note 77
APPENDIX A
DERIVATION OF OUTPUT RIPPLE CURRENT IN A
2-PHASE CIRCUIT
During the interval from DT to T as shown in Figure 1, the
high side switch in module 1 is off and the high side switch
of module 2 is on. The inductor current in module 1
decreases and the inductor current in module 2 increases.
The variations of these inductors are estimated to be:
− V (1 − D)T
∆IL1 = O
Lf
∆IL2 =
where
(A1)
( VIN − VO )(1 − D)T
Lf
D=
VO
VIN
Equation (A4) is derived based on the waveform shown in
Figure 1, where D is greater than 0.5. When D is smaller
than 0.5, one can easily derive the output ripple current as:
∆IO = ∆IL1 + ∆IL2 =
VO(1 − D)T 1 − 2D
1− D
Lf
(A5)
By combining equations (A4) and (A5), we can derive the
output ripple current for the 2-phase configuration as:
(A2)
∆IO = ∆IL1 + ∆IL2 =
2VO(1 − D)T 1 − 2D
Lf
1 − 2D + 1
(A6)
(A3)
The net output ripple current is the sum of these inductor
ripple currents:
∆IO = ∆IL1 + ∆IL2 =
AN77-16
VO(1 − D)T 1 − 2D
Lf
D
(A4)
Linear Technology Corporation
an77f LT/TP 0999 4K • PRINTED IN USA
1630 McCarthy Blvd., Milpitas, CA 95035-7417
(408)432-1900 ● FAX: (408) 434-0507 ● www.linear-tech.com
 LINEAR TECHNOLOGY CORPORATION 1999