STMICROELECTRONICS L6917B

L6917B
5 BIT PROGRAMMABLE DUAL-PHASE CONTROLLER
■
■
■
■
■
■
■
■
■
■
■
■
■
■
2 PHASE OPERATION WITH
SYNCRHONOUS RECTIFIER CONTROL
ULTRA FAST LOAD TRANSIENT RESPONSE
INTEGRATED HIGH CURRENT GATE
DRIVERS: UP TO 2A GATE CURRENT
TTL-COMPATIBLE 5 BIT PROGRAMMABLE
OUTPUT COMPLIANT WITH VRM 9.0
0.8% INTERNAL REFERENCE ACCURACY
10% ACTIVE CURRENT SHARING
ACCURACY
DIGITAL 2048 STEP SOFT-START
OVERVOLTAGE PROTECTION
OVERCURRENT PROTECTION REALIZED
USING THE LOWER MOSFET'S R dsON OR A
SENSE RESISTOR
300 kHz INTERNAL OSCILLATOR
OSCILLATOR EXTERNALLY ADJUSTABLE
UP TO 600kHz
POWER GOOD OUTPUT AND INHIBIT
FUNCTION
REMOTE SENSE BUFFER
PACKAGE: SO-28
SO-28
ORDERING NUMBERS:L6917BD
L6917BDTR (Tape & Reel)
DESCRIPTION
The device is a power supply controller specifically
designed to provide a high performance DC/DC conversion for high current microprocessors.
The device implements a dual-phase step-down controller with a 180° phase-shift between each phase.
A precise 5-bit digital to analog converter (DAC) allows adjusting the output voltage from 1.100V to
1.850V with 25mV binary steps.
The high precision internal reference assures the selected output voltage to be within ±0.8%. The high
peak current gate drive affords to have fast switching
to the external power mos providing low switching
losses.
The device assures a fast protection against load
over current and load over/under voltage. An internal
crowbar is provided turning on the low side mosfet if
an over-voltage is detected. In case of over-current,
the system works in Constant Current mode.
APPLICATIONS
■ POWER SUPPLY FOR SERVERS AND
WORKSTATIONS
■ POWER SUPPLY FOR HIGH CURRENT
MICROPROCESSORS
■ DISTRIBUTED DC-DC CONVERTERS
BLOCK DIAGRAM
ROSC / INH
SGND
VCCDR
PGOOD
2 PHASE
OSCILLATOR
PWM1
+
DIGITAL
SOFT START
CURRENT
CORRECTION
CH 1 OVER
CURRENT
LOGIC PWM
ADAPTIVE ANTI
CROSS-CONDUCTION
BOOT1
UGATE1
HS
PHASE1
LGATE1
LS
ISEN1
VCC
VCCDR
LOGIC
AND
PROTECTIONS
TOTAL
CURRENT
CURRENT
READING
+
PGNDS1
PGND
VID4
VID3
AVG
CURRENT
CH2 OVER
CURRENT
VID0
ISEN2
FBR
+
10k
IFB
10k
10k
CH 2 OVER
CURRENT
PWM2
ERROR
AMPLIFIER
REMOTE
BUFFER
LOGIC PWM
ADAPTIVE ANTI
CROSS-CONDUCTION
CH1 OVER
CURRENT
10k
FBG
PGNDS2
CURRENT
READING
<>
DAC
VID1
CURRENT
CORRECTION
VID2
LGATE2
LS
PHASE2
UGATE2
HS
Vcc
BOOT2
VSEN
September 2002
FB
COMP
Vcc
1/33
L6917B
ABSOLUTE MAXIMUM RATINGS
Symbol
Value
Unit
to PGND
15
V
Boot Voltage
15
V
15
V
-0.3 to Vcc+0.3
V
-0.3 to 7
V
26
V
Value
Unit
Thermal Resistance Junction to Ambient
60
°C/W
Maximum junction temperature
150
°C
Storage temperature range
-40 to 150
°C
Junction Temperature Range
-25 to 125
°C
2
W
Vcc, VCCDR
VBOOT-VPHASE
Parameter
VUGATE1-VPHASE1
VUGATE2-VPHASE2
LGATE1, PHASE1, LGATE2, PHASE2 to PGND
All other pins to PGND
Vphase
Sustainable Peak Voltage t < 20ns @ 600kHz
THERMAL DATA
Symbol
Rth j-amb
Tmax
Tstorage
Tj
PMAX
Parameter
Max power dissipation at Tamb = 25°C
PIN CONNECTION
LGATE1
1
28
PGND
VCCDR
2
27
LGATE2
PHASE1
3
26
PHASE2
UGATE1
4
25
UGATE2
BOOT1
5
24
BOOT2
VCC
6
23
PGOOD
GND
7
22
VID4
COMP
8
21
VID3
FB
9
20
VID2
VSEN
10
19
VID1
FBR
11
18
VID0
FBG
12
17
OSC / INH / FAULT
ISEN1
13
16
ISEN2
PGNDS1
14
15
PGNDS2
SO28
2/33
L6917B
ELECTRICAL CHARACTERISTICS
VCC = 12V ±10%, TJ = 0 to 70°C unless otherwise specified
Symbol
Parameter
Test Condition
Min
Typ
Max
Unit
Vcc SUPPLY CURRENT
Vcc supply current
HGATEx and LGATEx open
VCCDR=VBOOT=12V
7.5
10
12.5
mA
ICCDR
VCCDR supply current
LGATEx open; VCCDR=12V
2
3
4
mA
IBOOTx
Boot supply current
HGATEx open; PHASEx to PGND
VCC=VBOOT=12V
0.5
1
1.5
mA
Turn-On VCC threshold
VCC Rising; VCCDR=5V
7.8
9
10.2
V
Turn-Off VCC threshold
VCC Falling; VCCDR=5V
6.5
7.5
8.5
V
Turn-On VCCDR
Threshold
VCCDR Rising
VCC=12V
4.2
4.4
4.6
V
Turn-Off VCCDR
Threshold
VCCDR Falling
VCC=12V
4.0
4.2
4.4
V
OSC = OPEN
OSC = OPEN; Tj=0°C to 125°C
278
270
300
322
330
kHz
kHz
RT to GND=74kΩ
450
500
550
kHz
0.9
V
ICC
POWER-ON
OSCILLATOR/INHIBIT/FAULT
fOSC
Initial Accuracy
fOSC,Rosc Total Accuracy
INH
Inhibit threshold
ISINK=5mA
0.8
0.85
dMAX
Maximum duty cycle
OSC = OPEN
70
75
∆Vosc
Ramp Amplitude
1.8
2
2.2
V
FAULT
Voltage at pin OSC
OVP or UVP Active
4.75
5.0
5.25
V
Output Voltage
Accuracy
VID0, VID1, VID2, VID3, VID4
see Table1;
FBR = VOUT; FBG = GND
-0.8
-
0.8
%
VID pull-up Current
VIDx = GND
4
5
6
µA
VID pull-up Voltage
VIDx = OPEN
3.1
-
3.4
V
%
REFERENCE AND DAC
IDAC
ERROR AMPLIFIER
DC Gain
SR
80
dB
15
V/µs
DC Gain
1
V/V
Common Mode Rejection Ratio
40
dB
Slew-Rate
COMP=10pF
DIFFERENTIAL AMPLIFIER (REMOTE BUFFER)
CMRR
3/33
L6917B
ELECTRICAL CHARACTERISTICS (continued)
VCC = 12V ±10%, TJ = 0 to 70°C unless otherwise specified
Symbol
SR
Parameter
Test Condition
Input Offset
FBR=1.100V to1.850V;
FBG=GND
Slew Rate
VSEN=10pF
Min
Typ
-12
Max
Unit
12
mV
15
V/µs
DIFFERENTIAL CURRENT SENSING
45
50
55
µA
Bias Current
45
50
55
µA
Bias Current at
Over Current Threshold
80
85
90
µA
47.5
0
50
1
52.5
µA
µA
30
ns
IISEN1,
IISEN2
Bias Current
IPGNDSx
IISEN1,
IISEN2
IFB
Active Droop Current
Iload=0
Iload<0%
Iload=100%
GATE DRIVERS
High Side
Rise Time
VBOOTx-VPHASEx=10V;
CHGATEx to PHASEx=3.3nF
15
IHGATEx
High Side
Source Current
VBOOTx-VPHASEx=10V
2
RHGATEx
High Side
Sink Resistance
VBOOTx-VPHASEx=12V;
Low Side
Rise Time
ILGATEx
RLGATEx
tRISE
HGATE
tRISE
LGATE
A
2
2.5
Ω
VCCDR=10V;
CLGATEx to PGNDx=5.6nF
30
55
ns
Low Side
Source Current
VCCDR=10V
1.8
Low Side
Sink Resistance
VCCDR=12V
0.7
1.1
1.5
Ω
1.5
A
P GOOD and OVP/UVP PROTECTIONS
PGOOD
Upper Threshold
(VSEN/DACOUT)
VSEN Rising
108
112
116
%
PGOOD
Lower Threshold
(VSEN/DACOUT)
VSEN Falling
84
88
92
%
OVP
Over Voltage Threshold
(VSEN)
VSEN Rising
2.0
2.25
V
UVP
Under Voltage Trip
(VSEN/DACOUT)
VSEN Falling
56
60
64
%
PGOOD Voltage Low
IPGOOD = -4mA
0.3
0.4
0.5
V
VPGOOD
4/33
L6917B
Table 1. VID Settings
VID4
VID3
VID2
VID1
VID0
Output Voltage (V)
1
1
1
1
1
OUTPUT OFF
1
1
1
1
0
1.100
1
1
1
0
1
1.125
1
1
1
0
0
1.150
1
1
0
1
1
1.175
1
1
0
1
0
1.200
1
1
0
0
1
1.225
1
1
0
0
0
1.250
1
0
1
1
1
1.275
1
0
1
1
0
1.300
1
0
1
0
1
1.325
1
0
1
0
0
1.350
1
0
0
1
1
1.375
1
0
0
1
0
1.400
1
0
0
0
1
1.425
1
0
0
0
0
1.450
0
1
1
1
1
1.475
0
1
1
1
0
1.500
0
1
1
0
1
1.525
0
1
1
0
0
1.550
0
1
0
1
1
1.575
0
1
0
1
0
1.600
0
1
0
0
1
1.625
0
1
0
0
0
1.650
0
0
1
1
1
1.675
0
0
1
1
0
1.700
0
0
1
0
1
1.725
0
0
1
0
0
1.750
0
0
0
1
1
1.775
0
0
0
1
0
1.800
0
0
0
0
1
1.825
0
0
0
0
0
1.850
5/33
L6917B
PIN FUNCTION
N
Name
Description
1
LGATE1
Channel 1 low side gate driver output.
2
VCCDR
Mosfet driver supply. It can be varied from 5V to 12V.
3
PHASE1 This pin is connected to the source of the upper mosfet and provides the return path for the high
side driver of channel 1.
4
UGATE1 Channel 1 high side gate driver output.
5
BOOT1
6
VCC
Device supply voltage. The operative supply voltage is 12V.
7
GND
All the internal references are referred to this pin. Connect it to the PCB signal ground.
8
COMP
9
FB
This pin is connected to the error amplifier inverting input and is used to compensate the voltage
control feedback loop.
A current proportional to the sum of the current sensed in both channel is sourced from this pin
(50µA at full load, 70µA at the Over Current threshold). Connecting a resistor between this pin
and VSEN pin allows programming the droop effect.
10
VSEN
Connected to the output voltage it is able to manage Over & Under-voltage conditions and the
PGOOD signal. It is internally connected with the output of the Remote Sense Buffer for Remote
Sense of the regulated voltage.
If no Remote Sense is implemented, connect it directly to the regulated voltage in order to
manage OVP, UVP and PGOOD.
11
FBR
Remote sense buffer non-inverting input. It has to be connected to the positive side of the load to
perform a remote sense.
If no remote sense is implemented, connect directly to the output voltage (in this case connect
also the VSEN pin directly to the output regulated voltage).
12
FBG
Remote sense buffer inverting input. It has to be connected to the negative side of the load to
perform a remote sense.
Pull-down to ground if no remote sense is implemented.
13
ISEN1
Channel 1 bootstrap capacitor pin. Through this pin is supplied the high side driver and the upper
mosfet. Connect through a capacitor to the PHASE1 pin and through a diode to Vcc (cathode vs.
boot).
This pin is connected to the error amplifier output and is used to compensate the control
feedback loop.
Channel 1 current sense pin. The output current may be sensed across a sense resistor or
across the low-side mosfet RdsON. This pin has to be connected to the low-side mosfet drain or
to the sense resistor through a resistor Rg in order to program the positive current limit at 140%
as follow:
35 µA ⋅ R
IMA X = --------------------------g
R se nse
Where 35µA is the current offset information relative to the Over Current condition (offset at OC
threshold minus offset at zero load).
The net connecting the pin to the sense point must be routed as close as possible to the
PGNDS1 net in order to couple in common mode any picked-up noise.
14
PGNDS1 Channel 1 Power Ground sense pin. The net connecting the pin to the sense point (*) must be
routed as close as possible to the ISEN1 net in order to couple in common mode any picked-up
noise.
15
PGNDS2 Channel 2 Power Ground sense pin. The net connecting the pin to the sense point (*) must be
routed as close as possible to the ISEN2 net in order to couple in common mode any picked-up
noise.
(*) Through a resistor Rg.
6/33
L6917B
PIN FUNCTION (continued)
N
Name
Description
16
ISEN2
Channel 2 current sense pin. The output current may be sensed across a sense resistor or
across the low-side mosfet RdsON. This pin has to be connected to the low-side mosfet drain or
to the sense resistor through a resistor Rg in order to program the positive current limit at 140%
as follow:
35 µA ⋅ R
IMA X = --------------------------g
R se nse
Where 35µA is the current offset information relative to the Over Current condition (offset at OC
threshold minus offset at zero load).
The net connecting the pin to the sense point must be routed as close as possible to the
PGNDS2 net in order to couple in common mode any picked-up noise.
17
OSC/
INH/
FAULT
Oscillator switching frequency pin. Connecting an external resistor from this pin to GND, the
external frequency is increased according to the equation:
6
14.82 ⋅ 10
fS = 300KHz + ----------------------------R O SC ( KΩ )
Connecting a resistor from this pin to Vcc (12V), the switching frequency is reduced according to
the equation:
7
12.91 ⋅ 10
fS = 300KHz – ----------------------------R O SC ( KΩ )
If the pin is not connected, the switching frequency is 300KHz.
Forcing the pin to a voltage lower than 0.8V, the device stop operation and enter the inhibit state.
The pin is forced high when an over or under voltage is detected. This condition is latched; to
recover it is necessary turn off and on VCC.
18-22
VID4-0
Voltage IDentification pins. These input are internally pulled-up and TTL compatible. They are
used to program the output voltage as specified in Table 1 and to set the power good thresholds.
Connect to GND to program a ‘0’ while leave floating to program a ‘1’.
23
PGOOD
This pin is an open collector output and is pulled low if the output voltage is not within the above
specified thresholds.
If not used may be left floating.
24
BOOT2
Channel 2 bootstrap capacitor pin. Through this pin is supplied the high side driver and the upper
mosfet. Connect through a capacitor to the PHASE2 pin and through a diode to Vcc (cathode vs.
boot).
25
UGATE2 Channel 2 high side gate driver output.
26
PHASE2 This pin is connected to the source of the upper mosfet and provides the return path for the high
side driver of channel 2.
27
LGATE2
28
PGND
Channel 2 low side gate driver output.
Power ground pin. This pin is common to both sections and it must be connected through the
closest path to the low side mosfets source pins in order to reduce the noise injection into the
device.
7/33
L6917B
Device Description
The device is an integrated circuit realized in BCD technology. It provides complete control logic and protections
for a high performance dual-phase step-down DC-DC converter optimized for microprocessor power supply. It
is designed to drive N Channel MOSFETs in a dual-phase synchronous-rectified buck topology. A 180 deg
phase shift is provided between the two phases allowing reduction in the input capacitor current ripple, reducing
also the size and the losses. The output voltage of the converter can be precisely regulated, programming the
VID pins, from 1.100V to 1.850V with 25mV binary steps, with a maximum tolerance of ±0.8% over temperature
and line voltage variations. The device provides an average current-mode control with fast transient response.
It includes a 300kHz free-running oscillator adjustable up to 600kHz. The error amplifier features a 15V/µs slew
rate that permits high converter bandwidth for fast transient performances. Current information is read across
the lower mosfets rDSON or across a sense resistor in fully differential mode. The current information corrects
the PWM output in order to equalize the average current carried by each phase. Current sharing between the
two phases is then limited at ±10% over static and dynamic conditions. The device protects against over-current, with an OC threshold for each phase, entering in constant current mode. Since the current is read across
the low side mosfets, the constant current keeps constant the bottom of the inductors current triangular waveform. When an under voltage is detected the device latches and the FAULT pin is driven high. The device performs also over voltage protection that disable immediately the device turning ON the lower driver and driving
high the FAULT pin.
Oscillator
The device has been designed in order to operate an each phase at the same switching frequency of the internal
oscillator. So, input and output resulting frequency is doubled.
The switching frequency is internally fixed to 300kHz. The internal oscillator generates the triangular waveform
for the PWM charging and discharging with a constant current an internal capacitor. The current delivered to the
oscillator is typically 25µA and may be varied using an external resistor (R OSC) connected between OSC pin
and GND or Vcc. Since the OSC pin is maintained at fixed voltage (typ). 1.235V, the frequency is varied proportionally to the current sunk (forced) from (into) the pin considering the internal gain of 12KHz/µA.
In particular connecting it to GND the frequency is increased (current is sunk from the pin), while connecting ROSC
to Vcc=12V the frequency is reduced (current is forced into the pin), according to the following relationships:
6
1.237
kHz
14.82 ⋅ 10
R OS C vs. GND: f S = 300kH z + ------------------------------ ⋅ 12 ----------- = 300 kHz + -----------------------------( KΩ )
µA
R
( KΩ )
R
O SC
O SC
7
12 – 1.237
kHz
12.918 ⋅ 10
R OS C vs. 12V: f S = 300 kHz – ------------------------------ ⋅ 12 ----------- = 300kH z – -------------------------------( KΩ )
µA
R
( KΩ )
R
O SC
O SC
Note that forcing a 25µA current into this pin, the device stops switching because no current is delivered to the
oscillator.
Figure 1. ROSC vs. Switching Frequency
7000
1000
900
Rosc(KΩ) vs. GND
Rosc(KΩ) vs. 12V
6000
5000
4000
3000
2000
1000
700
600
500
400
300
200
100
0
0
100
200
Frequency (KHz)
8/33
800
300
0
300
400
500
600
700
800
Frequency (KHz)
900 1000
L6917B
Digital to Analog Converter
The built-in digital to analog converter allows the adjustment of the output voltage from 1.100V to 1.850V with
25mV as shown in the previous table 1. The internal reference is trimmed to ensure the precision of 0.8% and
a zero temperature coefficient around 70°C. The internal reference voltage for the regulation is programmed by
the voltage identification (VID) pins. These are TTL compatible inputs of an internal DAC that is realized by
means of a series of resistors providing a partition of the internal voltage reference. The VID code drives a multiplexer that selects a voltage on a precise point of the divider. The DAC output is delivered to an amplifier obtaining the VPROG voltage reference (i.e. the set-point of the error amplifier). Internal pull-ups are provided
(realized with a 5µA current generator up to 3.3V max); in this way, to program a logic "1" it is enough to leave
the pin floating, while to program a logic "0" it is enough to short the pin to GND. VID code “11111” programs
the NOCPU state: all mosfets are turned OFF and the condition is latched.
The voltage identification (VID) pin configuration also sets the power-good thresholds (PGOOD) and the overvoltage protection (OVP) thresholds.
Soft Start and INHIBIT
At start-up a ramp is generated increasing the loop reference from 0V to the final value programmed by VID in
2048 clock periods as shown in figure 2.
Before soft start, the lower power MOS are turned ON after that VCCDR reaches 2V (independently by Vcc value)
to discharge the output capacitor and to protect the load from high side mosfet failures. Once soft start begins,
the reference is increased; when it reaches the bottom of the oscillator triangular waveform (1V typ) also the
upper MOS begins to switch and the output voltage starts to increase with closed loop regulation.. At the end of
the digital soft start, the Power Good comparator is enabled and the PGOOD signal is then driven high (See fig.
2). The Under Voltage comparator enabled when the reference voltage reaches 0.8V.
The Soft-Start will not take place, if both V CC and VCCDR pins are not above their own turn-on thresholds. During normal operation, if any under-voltage is detected on one of the two supplies the device shuts down.
Forcing the OSC/INH/FAULT pin to a voltage lower than 0.8V the device enter in INHIBIT mode: all the power
mosfets are turned off until this condition is removed. When this pin is freed, the OSC/INH/FAULT pin reaches
the band-gap voltage and the soft start begins.
Figure 2. Soft Start
V IN =VCCDR
Turn ON threshold
2V
V LGATEx
t
V OUT
t
PGOOD
t
2048 Clock Cycles
Timing Diagram
t
Acquisition:
CH1 = PGOOD; CH2 = VOUT; CH4 = LGATEx
9/33
L6917B
Driver Section
The integrated high-current drivers allow using different types of power MOS (also multiple MOS to reduce the
RDSON), maintaining fast switching transition.
The drivers for the high-side mosfets use BOOTx pins for supply and PHASEx pins for return. The drivers for
the low-side mosfets use VCCDRV pin for supply and PGND pin for return. A minimum voltage of 4.6V at VCCDRV pin is required to start operations of the device.
The controller embodies a sophisticated anti-shoot-through system to minimize low side body diode conduction
time maintaining good efficiency saving the use of Schottky diodes. The dead time is reduced to few nanoseconds assuring that high-side and low-side mosfets are never switched on simultaneously: when the high-side
mosfet turns off, the voltage on its source begins to fall; when the voltage reaches 2V, the low-side mosfet gate
drive is applied with 30ns delay. When the low-side mosfet turns off, the voltage at LGATEx pin is sensed. When
it drops below 1V, the high-side mosfet gate drive is applied with a delay of 30ns. If the current flowing in the
inductor is negative, the source of high-side mosfet will never drop. To allow the turning on of the low-side mosfet even in this case, a watchdog controller is enabled: if the source of the high-side mosfet don't drop for more
than 240ns, the low side mosfet is switched on so allowing the negative current of the inductor to recirculate.
This mechanism allows the system to regulate even if the current is negative.
The BOOTx and VCCDR pins are separated from IC's power supply (VCC pin) as well as signal ground (SGND
pin) and power ground (PGND pin) in order to maximize the switching noise immunity. The separated supply
for the different drivers gives high flexibility in mosfet choice, allowing the use of logic-level mosfet. Several combination of supply can be chosen to optimize performance and efficiency of the application. Power conversion
is also flexible, 5V or 12V bus can be chosen freely.
The peak current is shown for both the upper and the lower driver of the two phases in figure 3. A 10nF capacitive load has been used. For the upper drivers, the source current is 1.9A while the sink current is 1.5A with
VBOOT-VPHASE = 12V; similarly, for the lower drivers, the source current is 2.4A while the sink current is 2A with
VCCDR = 12V.
Figure 3. Drivers peak current: High Side (left) and Low Side (right)
CH3 = HGATE1; CH4 = HGATE2
CH3 = LGATE1; CH4 = LGATE2
Current Reading and Over Current
The current flowing trough each phase is read using the voltage drop across the low side mosfets rDSON or
across a sense resistor (RSENSE) and internally converted into a current. The transconductance ratio is issued
by the external resistor Rg placed outside the chip between ISENx and PGNDSx pins toward the reading points.
The full differential current reading rejects noise and allows to place sensing element in different locations without affecting the measurement's accuracy. The current reading circuitry reads the current during the time in
10/33
L6917B
which the low-side mosfet is on (OFF Time). During this time, the reaction keeps the pin ISENx and PGNDSx
at the same voltage while during the time in which the reading circuitry is off, an internal clamp keeps these two
pins at the same voltage sinking from the ISENx pin the necessary current.
The proprietary current reading circuit allows a very precise and high bandwidth reading for both positive and
negative current. This circuit reproduces the current flowing through the sensing element using a high speed
Track & Hold transconductance amplifier. In particular, it reads the current during the second half of the OFF
time reducing noise injection into the device due to the mosfet turn-on (See fig. 4). Track time must be at least
200ns to make proper reading of the delivered current.
Figure 4. Current Reading Timing (Left) and Circuit (Right)
ILS1
LGATEX
Rg
ILS2
Rg
IPHASE
IISENx
Total current
information
RSENSE
ISENX
PGNDSX
50µA
Track & Hold
This circuit sources a constant 50µA current from the PGNDSx pin and keeps the pins ISENx and PGNDSx at
the same voltage. Referring to figure 4, the current that flows in the ISENx pin is then given by the following
equation:
R SENSE ⋅ I PHASE
I ISENx = 50µA + ---------------------------------------------- = 50µ A + I INFO x
Rg
Where RSENSE is an external sense resistor or the rds,on of the low side mosfet and Rg is the transconductance
resistor used between ISENx and PGNDSx pins toward the reading points; IPHASE is the current carried by each
phase and, in particular, the current measured in the middle of the oscillator period
The current information reproduced internally is represented by the second term of the previous equation as
follow:
R SENSE ⋅ I PHASE
I INFO x = ---------------------------------------------Rg
Since the current is read in differential mode, also negative current information is kept; this allow the device to
check for dangerous returning current between the two phases assuring the complete equalization between the
phase's currents.
From the current information of each phase, information about the total current delivered (IFB = IINFO1 + IINFO2)
and the average current for each phase (IAVG = (IINFO1 + IINFO2)/2 ) is taken. IINFOX is then compared to IAVG
to give the correction to the PWM output in order to equalize the current carried by the two phases.
The transconductance resistor Rg has to be designed in order to have current information of 25µA per phase
at full nominal load; the over current intervention threshold is set at 140% of the nominal (IINFOx = 35µA).
According to the above relationship, the limiting current (ILIM) for each phase, which has to be placed at one half
of the total delivered maximum current, results:
35 µA ⋅ R g
I L IM = --------------------------R SENSE
I LIM ⋅ R SENSE
Rg = ------------------------------------35µ A
An over current is detected when the current flowing into the sense element is greater than 140% of the nominal
11/33
L6917B
current (IINFOx>35µA): the device enters in Quasi-Constant-Current operation. The low-side mosfets stays ON
until IINFO becomes lower than 35µA skipping clock cycles. The high side mosfets can be turned ON with a TON
imposed by the control loop at the next available clock cycle and the device works in the usual way until another
OCP event is detected.
The device limits the bottom of the inductor current triangular waveform. So the average current delivered can
slightly increase also in Over Current condition since the current ripple increases. In fact, the ON time increases
due to the OFF time rise because of the current has to reach the 140% bottom. The worst-case condition is
when the duty cycle reaches its maximum value (d=75% internally limited). When this happens, the device
works in Constant Current and the output voltage decrease as the load increase. Crossing the UVP threshold
causes the device to latch (FAULT pin is driven high).
Figure 5 shows this working condition
Figure 5. Constant Current operation
Ipeak
Vout
Droop
effect
IMAX
140%
TonMAX
UVP
TonMAX
Inom
IOCP
IMAX
Iout
It can be observed that the peak current (Ipeak) is greater than the 140% but it can be determined as follow:
V IN – Vout M IN
Ipea k = 1.4 ⋅ I NOM + --------------------------------------- ⋅ To n M AX
L
Where INOM is the nominal current and Vout MIN is the minimum output voltage (VID-40% as explained below).
The device works in Constant-Current, and the output voltage decreases as the load increase, until the output
voltage reaches the under-voltage threshold (VoutMIN). When this threshold is crossed, all mosfets are turned
off, the FAULT pin is driven high and the device stops working. Cycle the power supply to restart operation.
The maximum average current during the Constant-Current behavior results:
Ip eak – 1.4 ⋅ I NOM
I M AX = 1.4 ⋅ I NOM + 2 ⋅ ------------------------------------------------2
In this particular situation, the switching frequency results reduced. The ON time is the maximum allowed (TonMAX) while the OFF time depends on the application:
Ipe ak – 1.4 ⋅ INOM
1
f = ------------------------------------------T O FF = L ⋅ ------------------------------------------------Vout
To n M AX + T O FF
Over current is set anyway when IINFOx reaches 35µA. The full load value is only a convention to work with convenient values for IFB. Since the OCP intervention threshold is fixed, to modify the percentage with respect to
the load value, it can be simply considered that, for example, to have on OCP threshold of 170%, this will correspond to IINFOx = 35µA (IFB = 70µA). The full load current will then correspond to IINFOx = 20.5µA (IFB = 41µA).
12/33
L6917B
Integrated Droop Function
The device uses a droop function to satisfy the requirements of high performance microprocessors, reducing
the size and the cost of the output capacitor.
This method "recovers" part of the drop due to the output capacitor ESR in the load transient, introducing a dependence of the output voltage on the load current
As shown in figure 6, the ESR drop is present in any case, but using the droop function the total deviation of the
output voltage is minimized. In practice the droop function introduces a static error (Vdroop in figure 6) proportional to the output current. Since the device has an average current mode regulation, the information about the
total current delivered is used to implement the Droop Function. This current (equal to the sum of both IINFOx)
is sourced from the FB pin. Connecting a resistor between this pin and Vout, the total current information flows
only in this resistor because the compensation network between FB and COMP has always a capacitor in series
(See fig. 7). The voltage regulated is then equal to:
VOUT = VID - RFB · IFB
Since IFB depends on the current information about the two phases, the output characteristic vs. load current is
given by:
R SENSE
V OUT = VID – R FB ⋅ ---------------------- ⋅ I OUT
Rg
Figure 6. Output transient response without (a) and with (b) the droop function
ESR DROP
ESR DROP
VMAX
VDROOP
VNOM
VMIN
(a)
(b)
Figure 7. Active Droop Function Circuit
RFB
COMP
To VOUT
FB
I FB
VPROG
The feedback current is equal to 50µA at nominal full load (IFB = IINFO1 + IINFO2) and 70µA at the OC threshold,
so the maximum output voltage deviation is equal to:
∆VFULL_POSITIVE_LOAD = +RFB · 50µA
∆VPOSITIVE_OC_THRESHOLD = +RFB · 70µA
Droop function is provided only for positive load; if negative load is applied, and then IINFOx < 0, no current is
sunk from the FB pin. The device regulates at the voltage programmed by the VID.
13/33
L6917B
Output Voltage Protection and Power Good
The output voltage is monitored by pin VSEN. If it is not within +12/-10% (typ.) of the programmed value, the
powergood output is forced low. Power good is an open drain output and it is enabled only after the soft start is
finished (2048 clock cycles after start-up).
The device provides over voltage protection; when the voltage sensed by the V SEN pin reaches 2.1V (typ.), the
controller permanently switches on both the low-side mosfets and switches off both the high-side mosfets in order to protect the CPU. The OSC/INH/FAULT pin is driven high (5V) and power supply (Vcc) turn off and on is
required to restart operations. The over Voltage percentage is set by the ratio between the OVP threshold (set
at 2.1V) and the reference programmed by VID.
2.1V
O VP[%] = ----------------------------------------------------------------------------- ⋅ 100
Refer ence Voltage ( VID )
Under voltage protection is also provided. If the output voltage drops below the 60% of the reference voltage for
more than one clock period the device turns off and the FAULT pin is driven high.
Both Over Voltage and Under Voltage are active also during soft start (Under Voltage after than Vout reaches
0.8V). During soft-start the reference voltage used to determine the OV and UV thresholds is the increasing voltage driven by the 2048 soft start digital counter.
Remote Voltage Sense
A remote sense buffer is integrated into the device to allow output voltage remote sense implementation without
any additional external components. In this way, the output voltage programmed is regulated between the remote buffer inputs compensating motherboard trace losses or connector losses if the device is used for a VRM
module. The very low offset amplifier senses the output voltage remotely through the pins FBR and FBG (FBR
is for the regulated voltage sense while FBG is for the ground sense) and reports this voltage internally at VSEN
pin with unity gain eliminating the errors.
If remote sense is not required, the output voltage is sensed by the VSEN pin connecting it directly to the output
voltage. In this case the FBG and FBR pins must be connected anyway to the regulated voltage.
Input Capacitor
The input capacitor is designed considering mainly the input rms current that depends on the duty cycle as reported in figure 8. Considering the dual-phase topology, the input rms current is highly reduced comparing with
a single phase operation.
Rms Current Normalized (IRMS/IOUT)
Figure 8. Input rms Current vs. Duty Cycle (D) and Driving Relationships
0.50
Single Phase
Dual Phase
0.25
I rms
IOUT

=  2
I
 OUT
 2
⋅
2D ⋅ (1 − 2D)
⋅ (2D - 1) ⋅ (2 − 2D)
if D < 0 .5
if D > 0.5
0.25
0.50
0.75
Duty Cycle (VOUT/VIN)
It can be observed that the input rms value is one half of the single-phase equivalent input current in the worst
case condition that happens for D = 0.25 and D = 0.75.
14/33
L6917B
The power dissipated by the input capacitance is then equal to:
P RM S = ESR ⋅ ( I RM S )
2
Input capacitor is designed in order to sustain the ripple relative to the maximum load duty cycle. To reach the
high rms value needed by the CPU power supply application and also to minimize components cost, the input
capacitance is realized by more than one physical capacitor. The equivalent rms current is simply the sum of
the single capacitor's rms current.
Input bulk capacitor must be equally divided between high-side drain mosfets and placed as close as possible
to reduce switching noise above all during load transient. Ceramic capacitor can also introduce benefits in high
frequency noise decoupling, noise generated by parasitic components along power path.
Output Capacitor
Since the microprocessors require a current variation beyond 50A doing load transients, with a slope in the
range of tenth A/µs, the output capacitor is a basic component for the fast response of the power supply.
Dual phase topology reduces the amount of output capacitance needed because of faster load transient response (switching frequency is doubled at the load connections). Current ripple cancellation due to the 180°
phase shift between the two phases also reduces requirements on the output ESR to sustain a specified voltage
ripple.
When a load transient is applied to the converter's output, for first few microseconds the current to the load is
supplied by the output capacitors. The controller recognizes immediately the load transient and increases the
duty cycle, but the current slope is limited by the inductor value.
The output voltage has a first drop due to the current variation inside the capacitor (neglecting the effect of the
ESL):
∆VOUT = ∆IOUT · ESR
A minimum capacitor value is required to sustain the current during the load transient without discharge it. The
voltage drop due to the output capacitor discharge is given by the following equation:
2
∆I OUT ⋅ L
∆V OUT = --------------------------------------------------------------------------------------------2 ⋅ C OUT ⋅ ( V INM IN ⋅ D M AX – V OUT )
Where DMAX is the maximum duty cycle value. The lower is the ESR, the lower is the output drop during load
transient and the lower is the output voltage static ripple.
Inductor design
The inductance value is defined by a compromise between the transient response time, the efficiency, the cost
and the size. The inductor has to be calculated to sustain the output and the input voltage variation to maintain
the ripple current ∆IL between 20% and 30% of the maximum output current. The inductance value can be calculated with this relationship:
V IN – V OUT V OUT
L = ------------------------------ ⋅ --------------V IN
fs ⋅ ∆I L
Where fSW is the switching frequency, VIN is the input voltage and V OUT is the output voltage.
Increasing the value of the inductance reduces the ripple current but, at the same time, reduces the converter
response time to a load transient. The response time is the time required by the inductor to change its current
from initial to final value. Since the inductor has not finished its charging time, the output current is supplied by
the output capacitors. Minimizing the response time can minimize the output capacitance required.
The response time to a load transient is different for the application or the removal of the load: if during the application of the load the inductor is charged by a voltage equal to the difference between the input and the output
15/33
L6917B
voltage, during the removal it is discharged only by the output voltage. The following expressions give approximate response time for ∆I load transient in case of enough fast compensation network response:
L ⋅ ∆I
t a pplic atio n = -----------------------------V IN – V OUT
L ⋅ ∆I
t rem ov al = --------------V OUT
The worst condition depends on the input voltage available and the output voltage selected. Anyway the worst
case is the response time after removal of the load with the minimum output voltage programmed and the maximum input voltage available.
Figure 9. Inductor ripple current vs Vout
9
L=1.5µH, Vin=12V
Inductor Ripple [A]
8
L=2µH,
Vin=12V
7
6
L=3µH,
Vin=12V
5
4
L=1.5µH,
Vin=5V
3
L=2µH,
Vin=5V
2
L=3µH, Vin=5V
1
0
0.5
1.5
2.5
3.5
Output Voltage [V]
MAIN CONTROL LOOP
The L6917B control loop is composed by the Current Sharing control loop and the Average Current Mode control loop. Each loop gives, with a proper gain, the correction to the PWM in order to minimize the error in its
regulation: the Current Sharing control loop equalize the currents in the inductors while the Average Current
Mode control loop fixes the output voltage equal to the reference programmed by VID. Figure 10 reports the
block diagram of the main control loop.
Figure 10. Main Control Loop Diagram
L1
+
PWM1
CURRENT
SHARING
DUTY CYCLE
CORRECTION
1/5
1/5
IINFO1
L2
+
PWM2
ERROR
AMPLIFIER
4/5
+
D02IN1392
REFERENCE
PROGRAMMED
BY VID
-
COMP
16/33
IINFO2
FB
ZF(S)
ZFB
CO
RO
L6917B
■
Current Sharing (CS) Control Loop
Active current sharing is implemented using the information from Tran conductance differential amplifier in an
average current mode control scheme. A current reference equal to the average of the read current (IAVG) is
internally built; the error between the read current and this reference is converted to a voltage with a proper gain
and it is used to adjust the duty cycle whose dominant value is set by the error amplifier at COMP pin (See fig.
11).
The current sharing control is a high bandwidth control loop allowing current sharing even during load transients.
The current sharing error is affected by the choice of external components; choose precise Rg resistor (±1% is
necessary) to sense the current. The current sharing error is internally dominated by the voltage offset of Tran
conductance differential amplifier; considering a voltage offset equal to 2mV across the sense resistor, the current reading error is given by the following equation:
∆I RE AD
2mV
-------------------- = ---------------------------------------R SENSE ⋅ I M AX
I M AX
Where ∆IREAD is the difference between one phase current and the ideal current (IMAX/2).
For Rsense = 4mΩ and Imax = 40A the current sharing error is equal to 2.5%, neglecting errors due to Rg and
Rsense mismatches.
Figure 11. Current Sharing Control Loop
+
L1
PWM1
CURRENT
SHARING
DUTY CYCLE
CORRECTION
1/5
1/5
+
PWM2
IINFO2
IINFO1
L2
COMP
VOUT
D02IN1393
■
Average Current Mode (ACM) Control Loop
The average current mode control loop is reported in figure 12. The current information IFB sourced by the FB
pin flows into RFB implementing the dependence of the output voltage from the read current.
The ACM control loop gain results (obtained opening the loop after the COMP pin):
PWM ⋅ Z F ( s ) ⋅ ( R DROOP + Z P ( s ) )
G LO O P ( s ) = -------------------------------------------------------------------------------------------------------------------ZF (s )
1
( Z P ( s ) + Z L ( s ) ) ⋅ --------------- +  1 + ------------  ⋅ R FB
A(s) 
A ( s )
Where:
R s en se
– R DROOP = ------------------- ⋅ R FB is the equivalent output resistance determined by the droop function;
Rg
– ZP(s) is the impedance resulting by the parallel of the output capacitor (and its ESR) and the applied
load Ro;
17/33
L6917B
– ZF(s) is the compensation network impedance;
– ZL(s) is the parallel of the two inductor impedance;
– A(s) is the error amplifier gain;
4 ∆V IN
– PWM = --- ⋅ ------------------- · is the ACM PWM transfer function where DVosc is the oscillator ramp amplitude
5 ∆V O SC
and has a typical value of 2V
Removing the dependence from the Error Amplifier gain, so assuming this gain high enough, the control loop
gain results:
ZF ( s)
V IN
 Rs Z P ( s )
4
G LO O P ( s ) = – --- ⋅ ------------------- ⋅ ------------------------------------ ⋅  -------- + ---------------
5 ∆V OS C Z P ( s ) + Z L ( s )  Rg R FB 
With further simplifications, it results:
Z F ( s ) R o + R DROOP
V IN
1 + s ⋅ Co ⋅ ( R DROOP //Ro + ESR )
4
G L OO P ( s ) = – --- ⋅ ------------------- ⋅ --------------- ⋅ -------------------------------------- ⋅ ---------------------------------------------------------------------------------------------------------------------------------5 ∆V O SC R FB
RL
RL
2
L
L
R o + ------s ⋅ C o ⋅ --- + s ⋅ --------------- + Co ⋅ ESR + Co ⋅ ------- + 1
2
2
2
2 ⋅ Ro
Considering now that in the application of interest it can be assumed that Ro>>RL; ESR<<Ro and
RDROOP<<Ro, it results:
V IN
1 + s ⋅ Co ⋅ ( R DROOP + ESR )
ZF (s )
4
G L OO P ( s ) = – --- ⋅ ------------------- ⋅ --------------- ⋅ ---------------------------------------------------------------------------------------------------------------------------------5 ∆V O SC R FB
RL
2
L
L
s ⋅ Co ⋅ --- + s ⋅ --------------- + C o ⋅ ESR + C o ⋅ ------- + 1
2
2
2 ⋅ Ro
The ACM control loop gain is designed to obtain a high DC gain to minimize static error and cross the 0dB axes
with a constant -20dB/dec slope with the desired crossover frequency ωT. Neglecting the effect of ZF(s), the
transfer function has one zero and two poles. Both the poles are fixed once the output filter is designed and the
zero is fixed by ESR and the Droop resistance. To obtain the desired shape an RF-CF series network is considered for the Z F(s) implementation. A zero at ωF=1/RFCF is then introduced together with an integrator. This integrator minimizes the static error while placing the zero in correspondence with the L-C resonance a simple 20dB/dec shape of the gain is assured (See Figure 12). In fact, considering the usual value for the output filter,
the LC resonance results to be at frequency lower than the above reported zero.
Figure 12. ACM Control Loop Gain Block Diagram (left) and Bode Diagram (right)
dB
IFB
ZF
CF
RF
GLOOP
RFB
VCOMP
K
REF
PWM
L/2
d•VIN
ZF(s)
VOUT
Cout
ESR
ωLC
Rout
1
4 VIN
K = --- ⋅ --------------- ⋅ ---------5 ∆Vo sc R FB
ωZ
ωT
ω
dB
Compensation network can be simply designed placing ωZ = ωLC and imposing the cross-over frequency ωT as
desired obtaining:
18/33
L6917B
L
C o ⋅ --2
C F = -------------------RF
R F B ⋅ ∆V O SC
5
L
R F = ---------------------------------- ⋅ ω T ⋅ -------------------------------------------------------- ⋅ --V IN
2 ⋅ ( R DROOP + ESR ) 4
LAYOUT GUIDELINES
Since the device manages control functions and high-current drivers, layout is one of the most important things
to consider when designing such high current applications.
A good layout solution can generate a benefit in lowering power dissipation on the power paths, reducing radiation and a proper connection between signal and power ground can optimise the performance of the control
loops.
Integrated power drivers reduce components count and interconnections between control functions and drivers,
reducing the board space.
Here below are listed the main points to focus on when starting a new layout and rules are suggested for a correct implementation.
■
Power Connections.
These are the connections where switching and continuous current flows from the input supply towards the load.
The first priority when placing components has to be reserved to this power section, minimizing the length of
each connection as much as possible.
To minimize noise and voltage spikes (EMI and losses) these interconnections must be a part of a power plane
and anyway realized by wide and thick copper traces. The critical components, i.e. the power transistors, must
be located as close as possible, together and to the controller. Considering that the "electrical" components reported in fig. 13 are composed by more than one "physical" component, a ground plane or "star" grounding connection is suggested to minimize effects due to multiple connections.
Figure 13. Power connections and related connections layout guidelines (same for both phases)
VIN
Rgate
VIN
BOOTx
CBOOTx
HS
HS
HGATEx
PHASEx
PHASEx
L
L
+VCC
Rgate
LGATEx
LS
COUT
D
CIN
PGNDx
a. PCB power and ground planes areas
VCC
LS
COUT
D
CIN
LOAD
SGND
LOAD
CVCC
b. PCB small signal components placement
Fig. 13a shows the details of the power connections involved and the current loops. The input capacitance
(CIN), or at least a portion of the total capacitance needed, has to be placed close to the power section in order
to eliminate the stray inductance generated by the copper traces. Low ESR and ESL capacitors (electrolytic or
Ceramic or both) are required.
■
Power Connections Related.
Fig.13b shows some small signal components placement, and how and where to mix signal and power ground
planes.
The distance from drivers and mosfet gates should be reduced as much as possible. Propagation delay times
19/33
L6917B
as well as for the voltage spikes generated by the distributed inductance along the copper traces are so minimized.
In fact, the further the mosfet is from the device, the longer is the interconnecting gate trace and as a consequence, the higher are the voltage spikes corresponding to the gate pwm rising and falling signals. Even if these
spikes are clamped by inherent internal diodes, propagation delays, noise and potential causes of instabilities
are introduced jeopardizing good system behavior. One important consequence is that the switching losses for
the high side mosfet are significantly increased.
For this reason, it is suggested to have the device oriented with the driver side towards the mosfets and the
GATEx and PHASEx traces walking together toward the high side mosfet in order to minimize distance (see fig
14). In addition, since the PHASEx pin is the return path for the high side driver, this pin must be connected
directly to the High Side mosfet Source pin to have a proper driving for this mosfet. For the LS mosfets, the return path is the PGND pin: it can be connected directly to the power ground plane (if implemented) or in the
same way to the LS mosfets Source pin. GATEx and PHASEx connections (and also PGND when no power
ground plane is implemented) must also be designed to handle current peaks in excess of 2A (30 mils wide is
suggested).
Gate resistors of few ohms help in reducing the power dissipated by the IC without compromising the system
efficiency.
Figure 14. Device orientation (left) and sense nets routing (right)
Towards HS mosfet
To LS mosfet
(or sense resistor)
(30 mils wide)
Towards LS mosfet
(30 mils wide)
Towards HS mosfet
(30 mils wide)
To LS mosfet
(or sense resistor)
To regulated output
The placement of other components is also important:
– The bootstrap capacitor must be placed as close as possible to the BOOTx and PHASEx pins to minimize the loop that is created.
– Decoupling capacitor from Vcc and SGND placed as close as possible to the involved pins.
– Decoupling capacitor from VCCDR and PGND placed as close as possible to those pins. This capacitor
sustains the peak currents requested by the low-side mosfet drivers.
– Refer to SGND all the sensible components such as frequency set-up resistor (when present) and also
the optional resistor from FB to GND used to give the positive droop effect.
– Connect SGND to PGND on the load side (output capacitor) to avoid undesirable load regulation effect
and to ensure the right precision to the regulation when the remote sense buffer is not used.
– An additional 100nF ceramic capacitor is suggested to place near HS mosfet drain. This helps in reducing noise.
– PHASE pin spikes. Since the HS mosfet switches in hard mode, heavy voltage spikes can be observed
on the PHASE pins. If these voltage spikes overcome the max breakdown voltage of the pin, the device
can absorb energy and it can cause damages. The voltage spikes must be limited by proper layout, the
use of gate resistors, Schottky diodes in parallel to the low side mosfets and/or snubber network on the
low side mosfets, to a value lower than 26V, for 20nSec, at Fosc of 600kHz max.
■ Current Sense Connections.
Remote Buffer: The input connections for this component must be routed as parallel nets from the FBG/FBR
20/33
L6917B
pins to the load in order to compensate losses along the output power traces and also to avoid the pick-up of
any common mode noise. Connecting these pins in points far from the load will cause a non-optimum load regulation, increasing output tolerance.
Current Reading: The Rg resistor has to be placed as close as possible to the ISENx and PGNDSx pins in
order to limit the noise injection into the device. The PCB traces connecting these resistors to the reading point
must be routed as parallel traces in order to avoid the pick-up of any common mode noise. It's also important
to avoid any offset in the measurement and to get a better precision, to connect the traces as close as possible
to the sensing elements, dedicated current sense resistor or low side mosfet Rdson.
Moreover, when using the low side mosfet RdsON as current sense element, the ISENx pin is practically connected to the PHASEx pin. DO NOT CONNECT THE PINS TOGETHER AND THEN TO THE HS SOURCE!
The device won't work properly because of the noise generated by the return of the high side driver. In this case
route two separate nets: connect the PHASEx pin to the HS Source (route together with HGATEx) with a wide
net (30 mils) and the ISENx pin to the LS Drain (route together with PGNDSx). Moreover, the PGNDSx pin is
always connected, through the Rg resistor, to the PGND: DO NOT CONNECT DIRECTLY TO THE PGND! In
this case, the device won't work properly. Route anyway to the LS mosfet source (together with ISENx net).
Right and wrong connections are reported in Figure 15.
Symmetrical layout is also suggested to avoid any unbalance between the two phases of the converter.
Figure 15. PCB layout connections for sense nets
CORRECT
NOT CORRECT
VIA to GND plane
To PHASE
connection
To LS Drain
and Source
To HS Gate
and Source
Wrong (left) and correct (right) connections for the current reading sensing nets.
APPLICATION EXAMPLES
The dual-pahse topology can be applied to several different applications ranging from CPU power supply (for
which the device has been designed) to standard high current DC-DC power supply. The application benefits
of all the advantages due to the dual-phase topology ranging from output ripple reduction to dynamic performance increase. After a general demo board overview, the following application examples will be illustrated:
– CPU Power Supply: 5 to 12 VIN; 1.7VOUT; 45A
– CPU Power Supply: 12VIN; VRM 9.0 Output; 50A
– High Current DC-DC: 12VIN; 3.3 to 5VOUTT; 35A
Demo Board Description
The demo board shows the operation of the device in a dual phase application. This evaluation board allows
output voltage adjustability (1.100V - 1.850V) through the switches S0-S4 and high output current capability.
The board has been laid out with the possibility to use up to two D2PACK mosfets for the low side switch in order
to give maximum flexibility in the mosfet's choice.
The four layers demo board's copper thickness is of 70µm in order to minimize conduction losses considering
the high current that the circuit is able to deliver.
Demo board schematic circuit is reported in Figure 16.
21/33
L6917B
Figure 16. Demo Board Schematic
Vin
JP6
DZ1
GNDin
JP2
JP1
VCCDR
Vcc
C5
D4
2
6
5
24
4
25
3
26
C7
BOOT1
C4
VCC
C8
GNDcc
UGATE1
Q2
UGATE2
Q4
C3
L2
PHASE2
R18
VoutCORE
R17
LGATE1
Q1
C6
R14
PHASE1
D1
D3
BOOT2
R15
L1
C9,C10 C11..C13
R10
R16
1
27
LGATE2
Q3
D2
R12
R13
ISEN1
Q1a
PGNDS1
14
R5
16
U1
R6
VID3
S2
VID2
S1
VID1
S0
VID0
R3
L6917B
22
R1
23
21
R4
PGOOD
PGOOD
VSEN
10
20
19
9
17
JP4
R8
R2
SGND
JP3
R7
FB
18
OSC / INH
GNDCORE
PGNDS2
15
PGND
VID4
R20
Q3a
28
S3
R19
ISEN2
13
S4
C14,
C23
C2
7
11
FBR
12
8
C1
JP5
R9
COMP
FBG
FBG
FBR
Several jumpers allow setting different configurations for the device: JP3, JP4 and JP5 allow configuring the
remote buffer as desired. Simply shorting JP4 and JP5 the remote buffer is enabled and it senses the output
voltage on-board; to implement a real remote sense, leave these jumpers open and connect the FBG and FBR
connectors on the demo board to the remote load. To avoid using the remote buffer, simply short all the jumpers
JP3, JP4 and JP5. Local sense through the R7 is used for the regulation.
The input can be configured in different ways using the jumpers JP1, JP2 and JP6; these jumpers control also
the mosfet driver supply voltage. Anyway, power conversion starts from VIN and the device is supplied from
VCC (See Figure 17).
Figure 17. Power supply configuration
To Vcc pin
To HS Drains (Power Input)
Vin
To BOOTx (HS Driver Supply)
JP6
GNDin
DZ1
JP2
JP1
Vcc
To VCCDR pin (LS Driver Supply)
GNDcc
Two main configurations can be distinguished: Single Supply (V CC = VIN = 12V) and Double Supply (VCC = 12V
VIN = 5V or different).
22/33
L6917B
– Single Supply: In this case JP6 has to be completely shorted. The device is supplied with the same rail
that is used for the conversion. With an additional zener diode DZ1 a lower voltage can be derived to
supply the mosfets driver if Logic level mosfet are used. In this case JP1 must be left open so that the
HS driver is supplied with VIN-VDZ1 through BOOTx and JP2 must be shorted to the left to use VIN or to
the right to use VIN-VDZ1 to supply the LS driver through VCCDR pin. Otherwise, JP1 must be shorted
and JP2 can be freely shorted in one of the two positions.
– Double Supply: In this case VCC supply directly the controller (12V) while VIN supplies the HS drains for
the power conversion. This last one can start indifferently from the 5V bus (Typ.) or from other buses
allowing maximum flexibility in the power conversion. Supply for the mosfet driver can be programmed
through the jumpers JP1, JP2 and JP6 as previously illustrated. JP6 selects now VCC or VIN depending
on the requirements.
Some examples are reported in the following Figures 18 and 19.
Figure 18. Jumpers configuration: Double Supply
Vcc = 12V
Vcc = 12V
HS Drains = 5V
HS Supply = 5V
Vin = 5V
GNDin
JP6
DZ1
GNDin
JP2
HS Drains = 5V
HS Supply = 12V
Vin = 5V
JP6
DZ1
JP2
JP1
JP1
VCCDR (LS Supply) = 5V
Vcc = 12V
VCCDR (LS Supply) = 12V
Vcc = 12V
GNDcc
GNDcc
(a) VCC = 12V; VBOOTx = VCCDR = VIN = 5V
(b) V CC = VBOOTx = VCCDR =12V; VIN = 5V
Figure 19. Jumpers configuration: Single Supply
Vcc = 12V
Vcc = 12V
HS Drains = 12V
Vin = 12V
HS Drains = 12V
HS Supply = 12V
Vin = 12V
HS Supply = 5.2V
GNDin
JP6
DZ1 6.8V
GNDin
JP2
Vcc = Open
JP6
DZ1
JP2
JP1
JP1
VCCDR (LS Supply) = 12V
VCCDR (LS Supply) = 12V
Vcc = Open
GNDcc
GNDcc
(a) VCC = VIN = VCCDR = 12V; VBOOTx = 5.2V
(b) VCC = VIN= VBOOTx = VCCDR = 12V
23/33
L6917B
PCB and Components Layouts
Figure 20. PCB and Components Layouts
Component Side
Internal PGND Plane
Figure 21. PCB and Components Layouts
Internal SGND Plane
24/33
Solder Side
L6917B
CPU Power Supply: 5 to 12VIN; 1.7VOUT; 45A
Considering the high slope for the load transient, a high switching frequency has to be used. In addition to fast
reaction, this helps in reducing output and input capacitor. Inductance value is also reduced.
A switching frequency of 200kHz for each phase is then considered allowing large bandwidth for the compensation network. It can be considered to use the 5V rail for the power conversion in order to allow compatibility
with standard ATX power supply.
– Current Reading Network and Over Current:
Since the maximum output current is IMAX = 45A, the over current threshold has been set to 46A (23A
per phase) in the worst case (max mosfet temperature). This because the device limits the valley of the
triangular ripple across the inductors. Considering to sense the output current across the low-side mosfet RdsON, STB90NF03L has 6.5mΩ max at 25°C that becomes 9.1mΩ considering the temperature
variation (+40%); the resulting Tran conductance resistor Rg has to be:
I MAX RdsO N
46 9.1m
Rg = ------------- ⋅ --------------------- = ------ ⋅ ------------- = 5.9kΩ
2
35 µ
2 35µ
(R3 to R6)
– Droop function Design:
Considering a voltage drop of 100mv at full load, the feedback resistor RFB has to be:
100mV
R FB = ------------------- = 1.43k Ω
70µ A
(R7)
– Inductor design:
Each phase has to deliver up to 22.5A; considering a current ripple of 5A (<25%), the resulting inductance value is:
1
12 – 1.7 1.7
Vin – Vo ut
d
L = ----------------------------- ⋅ ----------- = --------------------- ⋅ -------- ⋅ -------------------- = 1 µH
∆l
12 300000
5
Fsw
(L1, L2)
– Output Capacitor:
Five Rubycon MBZ (2200µF / 6.3V / 12mΩ max ESR) has been used implementing a resulting ESR of 2.4mΩ
resulting in an ESR voltage drop of 45A*2.4mΩ = 108mV after a 45A load transient.
– Compensation Network:
A voltage loop bandwidth of 20kHz is considered to let the device fast react after load transient.
The RF CF network results:
R F B ⋅ ∆V O SC 5
1µ
L
1.43k ⋅ 2 5
RF = ---------------------------------- ⋅ --- ⋅ ω T ⋅ ------------------------------------------------------- = ---------------------- ⋅ --- ⋅ 20k ⋅ 2π ⋅ ------------------------------------------------------------------- = 6200Ω (R8)
V IN
4
2 ⋅ ( R DRO OP + ESR )
12
4
9.1m

2 ⋅ ------------- ⋅ 1.43k + 2.4m
 5.9k

1µ
L
6 ⋅ 2200 µ ⋅ ------C o ⋅ --2
2
C F = -------------------- = ----------------------------------------- = 15 nF
6.2k
RF
(C2)
25/33
L6917B
Part List
R1
10k
SMD 0805
R2, R20
Not Mounted
SMD 0805
R3, R4, R5, R6
5.1k
1%
SMD 0805
R7
1.43k
1%
SMD 0805
R8
6.2k
SMD 0805
R10
82Ω
SMD 0805
R12 to R16, R19
2.2Ω
SMD 0805
R17, R18
0Ω
SMD 0805
C2
15n
SMD 0805
C3, C4
100n
SMD 0805
C5, C6, C7
1µ
Ceramic
SMD 1206
C8, C9, C10
10µ
Ceramic
SMD 1206
C11, C12, C13
1800µ / 16V
Rubycon MBZ
Radial 23x10.5
C19 to C24
2200µ / 6.3V
Rubycon MBZ
Radial 23x10.5
L1, L2
1µ
TO50 – 52B – 6 Turns
U1
L6917B
STMicroelectronics
SO28
Q1, Q3
STB90NF03L
STMicroelectronics
D2PACK
Q2, Q4
STB70NF03L
STMicroelectronics
D2PACK
D1, D2
STPS340U
STMicroelectronics
SMB
D3, D4
1N4148
STMicroelectronics
SOT23
System Efficiency
Figure 22 shows the demo board measured efficiency versus load current for different values of input voltage.
Mosfet temperature is always lower than 115 °C, at Tamb = 25°C.
Figure 22. Efficiency (fosc = 200kHz; Vout = 1.7V)
95
90
Efficiency [%]
85
80
75
70
65
60
Vin= 12V
Vin=5V
55
50
45
0
5
10
15
20
25
Output Current [A]
26/33
30
35
40
L6917B
CPU Power Supply: 12VIN - VRM 9.0 - 50A thermal
Figure 23 shows the device in a high current CPU core power supply solution.
The output voltage can be adjusted with binary step from 1.100V to 1.850V following VRM 9.0 specifications.
The demo board assembled with the following part list is capable to deliver up to 50A in open air without any
kind of airflow. Peak current can reach 60A without any limitations. For higher DC current, to avoid mosfet
change, airflow or heat sink are required.
Figure 23. CPU Power Supply Schematic
Vin
JP6
DZ1
GNDin
JP2
VCCDR
Vcc
C5
D4
2
6
5
24
4
25
VCC
C8
GNDcc
C7
BOOT1
UGATE1
C4
Q2
L1
UGATE2
Q4
C3
3
26
L2
PHASE2
R18
VoutCORE
R17
LGATE1
Q1
1
27
13
16
Q3
D2
R12
ISEN1
PGNDS1
14
R5
U1
VID3
S2
VID2
S1
VID1
S0
VID0
R3
L6917B
22
23
21
R1
10
20
PGOOD
VSEN
Ra
R7
FB
18
9
JP3
Ca
JP4
17
JP5
R8
R2
SGND
R4
PGOOD
19
OSC / INH
GNDCORE
PGNDS2
15
PGND
VID4
R20
Q3a
28
S3
R19
ISEN2
R6
S4
C14,
C23
LGATE2
R13
Q1a
Rosc
C6
R14
PHASE1
D1
D3
BOOT2
R15
To Vcc
C9,C10 C11..C13
R10
R16
JP1
C2
7
C1
R9
COMP
11
FBR
12
8
FBG
FBG
FBR
Part List
R1
10k
SMD 0805
R2, R9
Not Mounted
SMD 0805
R3, R4, R5, R6
3.3k
1%
SMD 0805
R7
3.6k
1%
SMD 0805
R8
3.3k
SMD 0805
R10
82
SMD 0805
R12 to R15
2.2
SMD 0805
R16, R17, R18
0
SMD 0805
Ra
1k
SMD 0805
Rosc
1.3M
C1
Not Mounted
1%
SMD 0805
SMD 0805
27/33
L6917B
Part List (continued)
C2
47n
SMD 0805
C3, C4
100n
Ceramic
SMD 0805
C5, C6, C7, C8
1µ
Ceramic
SMD 1206
C9, C10
10µ
Ceramic
SMD 1206
C11 to C13
1800µ/ 16V
Rubycon MBZ
Radial 10x10.5
C14 to C23
2200µ/ 6.3V
Rubycon MBZ
Radial 10x10.5
Ca
68n
L1, L2
0.5µ
77121 Core – 3 Turns
U1
L6917B
STMicroelectronics
SO28
Q1,Q1a, Q3,Q3a
SUB85N03-04P
Vishay - Siliconix
D2PACK
Q2, Q4
SUB70N03-09BP
Vishay - Siliconix
D2PACK
D1, D2
STPS340U
STMicroelectronics
SMB
D3, D4
1N4148
STMicroelectronics
SOT23
SMD 0805
Efficiency
Figure 24 showes the system efficiency for output current ranging form 5A up to 50A.
Figure 24. Efficiency (fosc = 200kHz; Vout = 1.7V)
89
Efficiency [%]
87
85
83
81
79
77
75
0
5
10
15
20
25
30
35
Output Current [A]
28/33
40
45
50
55
L6917B
Current Sharing
Figure 25 shows the current balancing between the two phases for different values of output current.
Figure 25.
Load Transient Response
Figure 26 shows the system response from 0 to 50A load transient. To obtain such a response, 5 additional
capacitors have been added to the output filter to reproduce the motherboard output filter. Noise can be further
reduced by adding ceramic decoupling capacitors.
Figure 26. 1.7V Output Voltage Ripple During 0 to 50A Load Transient
29/33
L6917B
High Current DC-DC: 12VIN - 3.3 (or 5V) OUT - 35A
Figure 27 shows the device in a high current server power supply application.
Adding an external resistor divider after the remote sense buffer gives the possibility to increase the regulated
voltage. Considering for example a divider by two (two equal resistors) the DAC range is doubled from 2.200V
to 3.700V with 50mV binary steps. The external resistor divider must be designed in order to give negligible effects to the remote buffer gain, this means that the resistors value must be much lower than the remote buffer
input resistance (20kΩ). In this way, it is possible to regulate the 3.3V and 2.5V rails from the 12V available from
the AC/DC converter. The 5V rail can be obtained with further modifications to the external divider. The regulator
assures all the advantages of the dual phase conversion especially in the 5V conversion where the duty cycle
is near the 50% and practically no ripple is present in the input capacitors.
The board is able to deliver up to 35A "thermal" at Tamb 25°C without airflow. Higher currents can be reached
for reasonable times considering the overall dynamic thermal capacitance.
Figure 27. Server power supply schematic
Vin
JP6
DZ1
GNDin
JP2
JP1
VCCDR
Vcc
C5
D4
VCC
2
6
5
24
4
25
C8
GNDcc
C7
BOOT1
Q2
Q4
C3
R14
PHASE1
L2
PHASE2
3
26
R18
Q1
1
27
Q3
D2
R12
R13
Q1a
PGNDS1
14
R5
16
U1
R6
VID3
S2
VID2
S1
VID1
S0
VID0
R3
L6917B
22
R1
R4
PGOOD
23
21
PGOOD
VSEN
10
20
19
JP3
R7
FB
18
9
OSC / INH
JP4
17
Rosc
GNDCORE
PGNDS2
15
PGND
VID4
R20
Q3a
28
S3
R19
ISEN2
13
S4
C14,
C23
LGATE2
ISEN1
To Vcc
VoutCORE
R17
LGATE1
D1
C6
UGATE2
R15
L1
D3
BOOT2
UGATE1
C4
C9,C10 C11..C13
R10
R16
JP5
R8
R2
C2
SGND
7
C1
R9
COMP
11
FBR
12
8
FBG
FBG
FBR
The following part list refers to the following application:
– Input Voltage: 12V;
– Output voltage: 3.3V;
– Oscilator frequency: 200kHz;
– Output voltage tolerance (over static and dynamic conditions): ±2.5%.
30/33
L6917B
Part List
R1
10k
SMD 0805
R2, R9
Not Mounted
SMD 0805
R3, R6
1.3k
1%
SMD 0805
R4, R5
390
1%
SMD 0805
R7
75
1%
SMD 0805
R8
750
SMD 0805
R10
82
SMD 0805
R12 to R15
2.2
SMD 0805
R16, R17, R18
0
SMD 0805
R19
300
1%
SMD 0805
R20
390
1%
SMD 0805
ROSC
1.3M
1%
SMD 0805
C1
Not Mounted
SMD 0805
C2
220n
SMD 0805
C3, C4
100n
Ceramic
SMD 0805
C5, C6, C7, C8
1µ
Ceramic
SMD 1206
C9, C10
10µ
Ceramic
SMD 1206
C11 to C13
100µ/ 20V
OSCON 20SA100M
Radial 10x10.5
C14,C16,C18,C20,C22
2200µ/16V
SANYO
Radial 10x23
L1, L2
2.8µ
77121 Core – 9 Turns
U1
L6917B
STMicroelectronics
SO28
Q1,Q1a,Q2, Q3,Q3a,Q4
STB90NF03L
STMicroelectronics
D2PACK
D1, D2
STPS340U
STMicroelectronics
SMB
D3, D4
1N4148
STMicroelectronics
SOT23
DZ1
Not Mounted
Minimelf
Figure 28. System Efficiency for a 12V/3.3V
Application (fosc = 200kHz)
Figure 29. Load Transient Response: 0A to 35A
@ 1A/µs
Efficiency [%]
93
92
91
90
89
88
0
5
10
15
20
25
30
35
40
Output Current [A]
31/33
L6917B
mm
DIM.
MIN.
TYP.
A
inch
MAX.
MIN.
TYP.
2.65
MAX.
0.104
a1
0.1
0.3
0.004
0.012
b
0.35
0.49
0.014
0.019
b1
0.23
0.32
0.009
0.013
C
0.5
c1
0.020
45° (typ.)
D
17.7
18.1
0.697
0.713
E
10
10.65
0.394
0.419
e
1.27
0.050
e3
16.51
0.65
F
7.4
7.6
0.291
0.299
L
0.4
1.27
0.016
0.050
S
32/33
OUTLINE AND
MECHANICAL DATA
8 ° (max.)
SO28
L6917B
Information furnished is believed to be accurate and reliable. However, STMicroelectronics assumes no responsibility for the consequences
of use of such information nor for any infringement of patents or other rights of third parties which may result from its use. No license is granted
by implication or otherwise under any patent or patent rights of STMicroelectronics. Specifications mentioned in this publication are subject
to change without notice. This publication supersedes and replaces all information previously supplied. STMicroelectronics products are not
authorized for use as critical components in life support devices or systems without express written approval of STMicroelectronics.
The ST logo is a registered trademark of STMicroelectronics
 2002 STMicroelectronics - All Rights Reserved
STMicroelectronics GROUP OF COMPANIES
Australia - Brazil - China - Finland - France - Germany - Hong Kong - India - Italy - Japan - Malaysia - Malta - Morocco - Singapore - Spain
- Sweden - Switzerland - United Kingdom - U.S.A.
http://www.st.com
33/33