NSC LM27262MTDX

LM27262
Intel CPU Core Voltage Regulator Controller for VRD10
Compatible PCs
General Description
The LM27262 is a versatile synchronous buck voltage regulator controller designed according to the Intel VRD10 Specification. It’s a fixed-frequency voltage-mode control PWM
with average current modulation of the reference voltage to
achieve the desired output impedance. This approach imparts a pseudo current mode behavior to the control loop.
The part provides the control for a voltage regulator consisting of either 2-, 3- or 4-phases to provide power to a desktop
CPU. Pulse-by-pulse phase current balancing ensures accurate current sharing. CPU core currents of over 90A can
be supported without requiring significant over design of the
power path. The LM27262 contains a precision 6-bit digitalto-analog converter (DAC) that uses a VID-code provided by
the CPU to program the desired CPU core voltage. The
regulator output voltage can be dynamically adjusted by
changing the VID-code “on the fly”.
The part is intended to provide all the specified functions laid
out in the VRD-10 specification, making it as simple as
possible for the user to realize a fully compliant CPU core
supply for Intel’s Pentium™4 and Prescott™ processors.
The LM27262 is available in a 48-lead TSSOP package and
in a 48-lead LLP package.
Features
n Compatible with “VRD10 Voltage Regulation
Specification” for Intel Pentium 4 and Prescott
Processors
n Supports Intel SpeedStep™ technology
(Geyserville-III™), which enables real-time dynamic
switching of the CPU core voltage and the CPU clock
frequency
n Uses external gate drivers (LM2724) for maximum
layout flexibility and noise immunity
n Fixed frequency PWM architecture
n Pin selectable internal or external voltage reference
n 0.5% core voltage set-point accuracy when using
external voltage reference
n 0.9% accurate internal bandgap reference
n 2-, 3- or 4-phase operation
n Out-of-phase switching reduces input ripple current,
thereby minimizing input capacitor requirements
n VID-code programmable output voltage range of
0.8375V through 1.6000V
n
n
n
n
n
n
n
n
n
n
User programmable, low loss, load line slope
User programmable Standard VID offset
User programmable VCORE slew rate
5V power rail Under Voltage Lock Out (UVLO)
Over Voltage Protection (OVP), Under Voltage
Protection (UVP), and Over Current Protection (OCP) to
defend against potentially catastrophic events
User programmable fault latch that can be used to
disable the regulator in the event a power system fault
Very fast transient response
VID-transition masked PWRGOOD output
True-differential current sensing for each phase ensures
accurate load current sharing
User programmable cycle-by-cycle current limit
Key Specifications
n Fast transient response to minimize output capacitor
requirements
n Shorter design cycles through the use of external gate
drivers for reduced power dissipation, ease of PCB
layout and reduced noise sensitivity
n Fully integrated solution. All VRD-10 control functions
provided by a single device
n Accurate current balancing eliminates need to over
design the power path
n Fixed frequency PWM minimizes EMI issues
n Low input capacitor requirements due to multi-phase
interleaving
n Fault latching allows use of smaller power path
components and minimizes the chance of damage to
the load in the event of a fault
n Only need to buy as many drivers as the design needs
while keeping only one controller in inventory
Applications
n Server and desktop computer CPU core power supplies
requiring a 2, 3- or 4-phase voltage regulator delivering
up to 100A
n Transportable notebook computers using desktop CPUs
n Low cost transportable notebook computers using 2- or
3-phase designs
Pentium™ is a trademark of Intel Corporation.
© 2005 National Semiconductor Corporation
DS200834
www.national.com
LM27262 Intel CPU Core Voltage Regulator Controller for VRD10 Compatible PCs
October 2004
LM27262
Typical Application Circuit
20083401
FIGURE 1.
www.national.com
2
LM27262
Connection Diagrams
Top View
Top View
20083403
48-Lead LLP
Order Number LM27262LQ
NS Package Number LQA48B
20083402
48-Lead TSSOP (MTD)
Order Number LM27262MTD
See NS Package Number MTD48
Ordering Information
Order Number
Package Drawing
LM27262MTD
MTD48
38 Units/Rail
LM27262MTDX
MTD48
1000 Units Tape and Reel
LM27262LQ
LQA48B
1000 Units Tape and Reel
LM27262LQX
LQA48B
4500 Units Tape and Reel
3
Supplied As
www.national.com
LM27262
Pin 32/27, DELAY: OVP, UVP and OCP latch-off delay
adjustment pin. A delay programming capacitor is connected
between this pin and ground. This pin disables UVP, OVP
and OCP latch-off when grounded to facilitate debugging
Pin 33/28, VCORE: CPU core voltage rail connection. This
pin is the OVP/UVP sense point.
Pin 34/29, COMP: output of error amplifier. Use for external
loop compensation connection
Pin Description
(All pin numbers referred to here correspond to the TSSOP/
LLP package)
Pin 1/44, PHASES: tri-level logic input: HIGH logic level
switches controller into 2-phase operation mode, phases A
and C active. LOW logic level activates 3-phase operation,
phases A, B and C active, OPEN (floating) input activates
4-phase operation
Pin 2/45, IREF: connect a 1% resistor to ground to program
a precision current source for a standard offset voltage,
typically –25mV, across a resistor connected between
VPROG and VSTDOS pins. Recommended current value is
approximately 80µA. Typical resistor value is R = 1.4V /
80µA equals 17.4k.
Pin 3/46, VSTDOS: input; VVSTDOS = VVPROG – VOS. This
pin allows setting a programmable offset voltage (typically
25mV). The offset is programmed via an external 1% resistor
connected between the VPROG and VSTDOS pins. The
offset is the Pin 2 current multiplied by this offset programming resistor.
Pin 35/30, VFB: input of error amplifier. Use for external loop
compensation connection Pin 36/31, VILD: phase D current
sense resistor low-side connection input
Pin 37/32, VIHD: phase D current sense resistor high-side
connection input
Pin 38/33, VILC: phase C current sense resistor low-side
connection input
Pin 39/34, VIHC: phase C current sense resistor high-side
connection input
Pin 40/35, VILB: phase B current sense resistor low-side
connection input
Pin 41/36, VIHB: phase B current sense resistor high-side
connection input
Pin 42/37, VILA: phase A current sense resistor low-side
connection input
Pin 43/38, VIHA: phase A current sense resistor high-side
connection input
Pin 44/39, VIDPGD: VID Power Good Delayed output. Outputs a VID_PWRGD signal that is delayed approximately
2msec after receiving an externally supplied active high
signal to VRON. Pin VIDPGD should be connected to the
system’s VID_PWRGD input. This delay ensures that Vcore
will power on only after the 6 VID bit signals have settled.
The LM27262 is only enabled after the delay has timed out.
Pin 45/40, VIDSLEW: connect a resistor between this pin
and the SOFTCAP pin to program VCORE slew rates for
VID transitions
Pin 4/47, VPROG: output used for programming a standard
offset. Connect a 1% resistor between VPROG and VSTDOS. VPROG output voltage is the buffered internal DAC
output.
Pin 5/48: No connect pin
Pin 6/1, SLADJ1: load line slope adjustment via external
resistor divider
Pin 7/2, SLADJ2: load line slope adjustment via external
resistor divider
Pin 8 – 13/3-8, VID0-VID5: voltage identification code inputs
Pin 14/9, VDAC: buffered output of onboard DAC. Voltage
determined by VID code.
Pin 15-18/10-13: no connect pin
Pin 19/14, CLIMADJ: output current limit adjustment input
for one phase. For 4-phase operation the current limit is 4x
the single phase current limit, for 3-phase operation it is 3x
the single phase limit, and 2x the single phase current for
2-phase operation
Pin 20/15, REFINT: internal/external voltage reference selection logic input. When logic high, selects internal reference
Pin 21/16, VCC5V: 5V power supply input to the part.
Should be decoupled to GND pin with a 1uF capacitor.
Pin 22/17, VREF: internal voltage reference output or external voltage reference input depending on REFINT input logic
state
Pin 23/18, GND: the chip ground pin. Use for 5V supply
ground connection, Make a single-point ground connection
at this pin.
Pin 24 – 27/19-22: no connect pins
Pin 28-31/23-26: DRIVED-DRIVEA: PWM logic level outputs for phases D through A. Not short-circuit protected.
www.national.com
Pin 46/41, SOFTCAP: soft start/soft stop capacitor connection; this output sources charging current to the softstart
capacitor at power on. An internal 50k resistor discharges
the softstart capacitor during power off
Pin 47/42, PWRGD: power good output, open drain, active
high
Pin 48/43, VRON: logic input that turns the switching regulator on and off. If VCC5V is present when the LM27262 is
shutdown then the DRIVEx outputs are active low. VRON
has a 2msec assertion delay. When VRON is de-asserted,
the VID DAC latches the latest VID code and executes
soft-stop. There is no de-assertion delay on VRON.
LLP DAP, SUB: die substrate. The exposed die attach
should be connected to ground potentful.
4
Processor Pins (0 = low, 1 = high)
VID4
VID3
VID2
VID1
VID0
VOUT (V)
VID5
VID4
VID3
VID2
VID1
VID0
VOUT (V)
0
0
1
0
1
0
0.8375
0
1
1
0
1
0
1.2125
1
0
1
0
0
1
0.8500
1
1
1
0
0
1
1.2250
0
0
1
0
0
1
0.8625
0
1
1
0
0
1
1.2375
1
0
1
0
0
0
0.8750
1
1
1
0
0
0
1.2500
0
0
1
0
0
0
0.8875
0
1
1
0
0
0
1.2625
1
0
0
1
1
1
0.9000
1
1
0
1
1
1
1.2750
0
0
0
1
1
1
0.9125
0
1
0
1
1
1
1.2875
1
0
0
1
1
0
0.9250
1
1
0
1
1
0
1.3000
0
0
0
1
1
0
0.9375
0
1
0
1
1
0
1.3125
1
0
0
1
0
1
0.9500
1
1
0
1
0
1
1.3250
0
0
0
1
0
1
0.9625
0
1
0
1
0
1
1.3375
1
0
0
1
0
0
0.9750
1
1
0
1
0
0
1.3500
0
0
0
1
0
0
0.9875
0
1
0
1
0
0
1.3625
1
0
0
0
1
1
1.0000
1
1
0
0
1
1
1.3750
0
0
0
0
1
1
1.0125
0
1
0
0
1
1
1.3875
1
0
0
0
1
0
1.0250
1
1
0
0
1
0
1.4000
0
0
0
0
1
0
1.0375
0
1
0
0
1
0
1.4125
1
0
0
0
0
1
1.0500
1
1
0
0
0
1
1.4250
0
0
0
0
0
1
1.0625
0
1
0
0
0
1
1.4375
1
0
0
0
0
0
1.0750
1
1
0
0
0
0
1.4500
0
0
0
0
0
0
1.0875
0
1
0
0
0
0
1.4625
1
1
1
1
1
1
OFF
1
0
1
1
1
1
1.4750
0
1
1
1
1
1
OFF
0
0
1
1
1
1
1.4875
1
1
1
1
1
0
1.1000
1
0
1
1
1
0
1.5000
0
1
1
1
1
0
1.1125
0
0
1
1
1
0
1.5125
1
1
1
1
0
1
1.1250
1
0
1
1
0
1
1.5250
0
1
1
1
0
1
1.1375
0
0
1
1
0
1
1.5375
1
1
1
1
0
0
1.1500
1
0
1
1
0
0
1.5500
0
1
1
1
0
0
1.1625
0
0
1
1
0
0
1.5625
1
1
1
0
1
1
1.1750
1
0
1
0
1
1
1.5750
0
1
1
0
1
1
1.1875
0
0
1
0
1
1
1.5875
1
1
1
0
1
0
1.2000
1
0
1
0
1
0
1.6000
FIGURE 2. VID Code vs DAC Output
5
www.national.com
LM27262
Processor Pins (0 = low, 1 = high)
VID5
LM27262
Absolute Maximum Ratings (Notes 1, 4)
Junction Temperature
If Military/Aerospace specified devices are required,
please contact the National Semiconductor Sales Office/
Distributors for availability and specifications.
Minimum ESD Rating
(Note 4)
Human Body Model
Machine Model
100pF
1.5 kΩ
Soldering Dwell Time,
Temperature (Note 3)
Wave
Infrared
Vapor Phase
4sec, 260˚C
10sec, 240˚C
75sec, 219˚C
VCC5V
-0.3V to 7V
PHASES, IREF, VSTDOS,
VPROG,
SLADJ1, SLADJ2, VREF,
REFINT,
CLIMADJ, VCORE, COMP,
VFB,
VILx, VIHx, SOFTCAP,
PWRGD,
VIDPGD, VID0-VID5,
VRON
-0.3V to VCC5V + 0.3V
Differential Voltage (VILx –
VIHx)
1V
± 2kV
Operating Ratings (Note 1)
Ambient Storage Temp.
Range
VCC5V
4.65V to 5.5V
Junction Temperature
(Note 2)
0˚C to +110˚C
-65˚C to +150˚C
Electrical Characteristics (Note 5),(Note 6)
Symbol
-20˚C to +150˚C
Parameter
VCC5V = 5V unless otherwise specified.
Conditions
Min
Typ
Max
IQ5V
Quiescent VCC5V
current
VCC5V = 5.5V, not switching
8.4
10.5
IQ5V
Quiescent VCC5V
current
VCC5V = 5.5V, 4-phase switching
8.5
11
ISD5V
Shutdown VCC5V
current
VCC5V = 5.5V, VRON = Low
0.1
10
TSD
Thermal Shutdown
Threshold
Rising temperature
165
TSDH
Thermal Shutdown
Threshold Hysteresis
Unit
mA
mA
µA
˚C
10
˚C
UNDER VOLTAGE LOCKOUT
V5UVLO
V5UVLOH
VCC5V UVLO
THreshold
Rising Edge
4.15
4.3
4.6
Falling Edge
3.85
4.05
4.35
VCC5V UVLO
Hysteresis
0.25
V
V
DAC: VID0-5, VDAC
VID0-5 Inputs Logic
LOW
0.4
VID0-5 Inputs Logic
HIGH
0.8
V
VID0-5 Low-to-High
Threshold
0.7
VID0-5 Threshold
Hysteresis
0.2
VID0-5 Inputs Internal
Pull-up Current
VID0-5 = Low
DAC Output Voltage
Programming
Resolution
Per VID code table, Measured at VPROG
5
V
V
20
µA
12.5
mV
DAC Accuracy measured at VPROG pin over
-5˚C < TJ < 110˚C
www.national.com
V
6
Symbol
Parameter
Conditions
Min
DAC Voltage Accuracy VID codes from 1.5625V to 1.6000V
with Internal VREF
VID codes from 0.8375V to 1.1000V
VID codes from 1.1125V to 1.5500V
DAC Voltage Accuracy VID codes: 1.5500V & 1.6000V
with External VREF =
1.225V
VID codes: 0.8375V & 1.0000V
VID codes: 1.1000V, 1.2000, 1.3000, 1.4000 &
1.5000
-0.6
-0.8
Typ
± 0.9
± 0.9
± 0.75
± 0.15
(Continued)
Max
Unit
%
%
%
0.75
0.8
%
-1.0
-1.2
-0.15
1.0
1.2
%
-0.5
-0.8
± 0.1
0.5
0.8
%
1.210
1.235
1.260
REFERENCE VOLTAGE
Internal Reference
Voltage
VREF Output Load
Regulation
IVREF from 0 to 50 µA
-2.5
V
mV
VREF Line Regulation VCC5V = (4.65V to 5.5V), IVREF = 50uA
± 0.2
mV
External VREF Voltage REFINT=LOW, see LM27201 Electrical Specs.
1.225
V
External VREF
Compensating Offset
(applied internally)
REFINT=LOW
10
mV
STANDARD OFFSET PROGRAMMING INPUTS: VPROG, VSTDOS, IREF
IREF Output Voltage
100 µA Load Current
VOS
(VSTDOS-VPROG)
IREF = 80.4 µA, ROS = 309
1.400
V
-25
mV
LOAD LINE SLOPE ADJUSTMENT (SLADJ), CURRENT LIMIT (CLIMADJ)
SLADJ2 Input Source
Current
SLADJ2 connected to GND
0.07
CLIMADJ Input Source VCLIMADJ = VREF
Current
2
Load Line Slope (LLS) SLADJ divider 41.2k/8.45k, RSENSE = 2mΩ
LLS Maximum Error
µA
Nominal LLS = -1.3mV/A, ISLADJ1 = 50 µA,
Isense diff. input = 20mV; not switching
µA
-1.3
mΩ
± 0.06
mΩ
CURRENT SENSE LINES VILx AND VIHx
± 0.5
Current Sense
Amplifier Input Offset
Voltage
Common Mode Voltage = 1.30V
Current Sense
Differential Voltage
Range
Over full VID range
Current Sense Input
Source Current
VIHx = VILx = 100mV
14
Differential Input
Resistance
Common Mode Voltage = 1.3V
1
mV
0-100
mV
25
µA
kΩ
CORE VOLTAGE ERROR AMPLIFIER
VCORE Input Bias
Current
VFB = 1.4V
0.3
COMP Output Sink
Current
Vcomp = 2.5V VFB = 5V
100
COMP Output Source
Current
Vcomp = 4V VFB = 0V
0.5
VFB Input Bias Source VFB = 1.4V VDAC = 1.3V
Current
µA
µA
mA
1.0
7
µA
www.national.com
LM27262
Electrical Characteristics (Note 5),(Note 6) VCC5V = 5V unless otherwise specified.
LM27262
Electrical Characteristics (Note 5),(Note 6) VCC5V = 5V unless otherwise specified.
Symbol
Parameter
Conditions
Min
Typ
Error Amp Input Offset VDAC = 1.3V
Voltage
±1
Error Amplifier Open
loop DC Gain
730
VFB = 1.3V, Comp Pin open
(Continued)
Max
Unit
mV
V/V
OSCILLATOR
Active Mode Switching
Frequency
235
300
345
kHz
Minimum On-Time
120
ns
Maximum Duty Cycle
75
%
SOFT START AND SOFT STOP (SOFTCAP), VIDSLEW
SOFTCAP Charge
Current
SOFTCAP Discharge
Resistor
2.2
Measured from SOFTCAP pin to GND pin
3.2
4.4
50
µA
kΩ
DELAY FUNCTION
DELAY Source/Charge VDELAY = 0V
Current
9
12
DELAY Discharge
Current
VDELAY = 5V
1
DELAY Threshold
Voltage
VDELAY rising
1.4
15
µA
mA
V
PWRGOOD, UVP, OVP FAULT LATCHING THRESHOLD
VPGH%
VPGL%
PWRGD OVP
Threshold
Difference of VCORE pin above EA int.
reference voltage measured at VFB pin in test
mode; VID = 110110 = 1.3V, VOS = 25mV
PWRGD UVP
Threshold
Percentage of VCORE pin below EA int.
reference voltage measured at VFB pin in test
mode; VID = 110110 = 1.3V, VOS = 25mV
PWRGD Output Low
Voltage
PWRGD sinking 4mA (open drain)
PWRGD Leakage
Current
PWRGD pulled up to 5.5V
VCORE OVP Latch
Threshold
Difference of VCORE pin above EA int.
reference voltage measured at VFB pin in test
mode; VID = 110110 = 1.3V, VOS = 25mV
VCORE UVP Latch
Threshold
Percentage of VCORE pin below EA int.
reference voltage measured at VFB pin in test
mode; VID = 110110 = 1.3V, VOS = 25mV
0.23
V
86
89
92
%
0.15
1
V
10
µA
0.23
V
86
89
92
%
VRON
Low Logic Level Input
Voltage
0.3
High Logic Level Input
Voltage
1.2
V
Low-to-High Voltage
Threshold
0.95
-
VRON Threshold
Hysteresis
0.1
-
VRON Leakage Current VVRON = 3.3V
5
20
VVRON = GND
50
Delay of power on after VRON transition
(SOFTCAP = 0.01µF)
2
VRON Assertion Delay
www.national.com
8
V
V
V
µA
nA
msec
Symbol
Parameter
Conditions
Min
(Continued)
Typ
VRON De-assertion
Delay
Delay of VRON falling edge by the internal logic
50
REFINT Leakage
Current
REFINT = 3.3V or GND
±2
Max
Unit
nsec
REFINT
Logic Low Input
Voltage
0.5
Logic High Input
Voltage
3.0
µA
V
V
PHASES
PHASES Leakage
Current
VPHASES =0V
VPHASES = 3.3V
- 25
µA
4
µA
PHASES Open Circuit 10MΩ to Ground
Voltage
2.8
Logic Low Max Input
Voltage
0.2
Logic High Min Input
Voltage
Relative to Vcc (5V nominal)
-0.2
Output Source Current is 10 mA
3.5
Output Source Current is 0 mA (no load)
4.5
Output Sink Current is 10 mA
0.3
V
V
V
LOGIC OUTPUTS: DRIVEx
Output High Voltage
Output Low Voltage
Low-to-High Transition 10% to 90% of VCC5V,
Time
CLOAD = 50pF
20
High-to-Low Transition 90% to 10% of VCC5V,
Time
CLOAD = 50pF
20
V
V
ns
ns
VIDPGD
VIDPGD Output Max
Low Voltage
VIDPGD sinking 4mA (open drain)
0.17
0.25
VIDPGD Leakage
Current
VIDPGD = 5.5V
1
10
VIDPGD Assertion
Delay
VRON assertion to VIDPGD assertion
(SOFTCAP = 0.01µF)
2
VIDPGD De-assertion
Delay
VRON de-assertion to VIDPGD de-assertion
50
V
µA
ms
ns
Note 1: Absolute maximum ratings indicate limits beyond which damage to the device may occur. Operating Ratings are conditions under which operation of the
device is guaranteed. For guaranteed performance limits and associated test conditions, see the Electrical Characteristics table.
Note 2: The maximum allowable power dissipation is calculated by using PDmax = (TJMAX - TA) /θJA , where TJMAX is the maximum junction temperature, TA is the
ambient temperature, and θJA is the junction-to-ambient thermal resistance of the specified package. The 1.56W rating results from using 150˚C, 25˚C, and 80˚C/W
for TJMAX, TA, and θJA respectively. The θJA of 90˚C/W represents the worst-case condition with no heat sinking of the 48-Pin TSSOP. Heat sinking allows the safe
dissipation of more power. The Absolute Maximum power dissipation should be de-rated by 12.5mW per ˚C above 25˚C ambient. The LM27262 actively limits its
junction temperature to about 165˚C.
Note 3: For detailed information on soldering plastic small-outline packages, refer to the Packaging Databook available from National Semiconductor Corporation.
Note 4: For testing purposes, ESD was applied using the human-body model, a 100pF capacitor discharged through a 1.5kΩ resistor.
Note 5: All limits are guaranteed at room temperature (standard face type) and at temperature extremes (bold face type). All room temperature limits are 100%
production tested. All limits at temperature extremes are guaranteed via correlation using Statistical Quality Control (SQC) methods. All limits are used to calculate
Average Outgoing Quality Level (AOQL).
Note 6: A “typical” specification is the center of characterization data distribution taken with TA = TJ = 25˚C. Typical data are not guaranteed.
9
www.national.com
LM27262
Electrical Characteristics (Note 5),(Note 6) VCC5V = 5V unless otherwise specified.
www.national.com
10
Block Diagram
20083404
LM27262
LM27262
Typical Performance Characteristics
LM27262 Efficiency vs ICORE
LM27262 VCORE vs ICORE
20083406
20083405
Soft-Start
Soft-Start
20083407
20083408
Soft-Stop at ICORE = 0A
Soft-Stop at ICORE = 30A
20083409
20083410
11
www.national.com
LM27262
Typical Performance Characteristics
(Continued)
Load Transient 24A to 75A
Load Transient 75A to 24A
20083413
20083414
Load Transient 24A to 75A to 24A
Over Current Protection Threshold vs Temperature
20083415
20083416
VDAC vs Temperature
External Ref, DAC Trim @ 1.525V
VDAC vs Temperature
External Ref, DAC Trim @ 1.325V
20083418
20083417
www.national.com
12
LM27262
Typical Performance Characteristics
(Continued)
VDAC vs Temperature
Internal Ref, DAC Trim @ 1.325V
External VREF vs Temperature (LM27201)
20083420
20083419
Switching Frequency Percent Change vs Temperature
Pin IREF Voltage vs Temperature
20083421
20083422
13
www.national.com
LM27262
error amplifier, the output of which drives an adjustable
current source, I2. Current source I1 provides a continuous
current for charging the internal ramp capacitor, CRAMP,
while I2 makes slight adjustments to this charging current.
The resulting ramp voltage is ultimately compared to the
voltage loop’s error amplifier output to control the regulator’s
pulse width. When the PWM comparator trips, it also turns
on the ramp generator’s reset transistor and dumps the ramp
capacitor. Amplifiers A3 and A5 perform the same function
for Phase C by controlling an identical ramp generator. In
summary, the slope of the PWM generator’s ramp signal is
adjusted as required to keep the phase currents balanced.
For instance, if the phase A current is a bit too high compared to the average phase current, the slope of the PWM
ramp for this phase is increased slightly. This tends to turn
the phase off a bit early and reduce its output current.
Operation Descriptions
GENERAL
The LM27262 is a selectable 2-, 3-, or 4-phase step down
switching regulator controller. It’s a fixed-frequency, voltagemode control PWM with user programmable average current
modulation of the reference voltage. This approach imparts a
pseudo current mode behavior to the control loop as well as
load line shaping for improved dynamic performance. The
individual phase currents are continuously monitored and
the duty cycles of each phase are adjusted as necessary so
that the phase currents are all kept equal. The MOSFET
drivers are contained in separate driver chips. This offers
several advantages. From a cost standpoint, the largest
amount of die area in most controllers is used for the drivers.
As such, with external drivers, only the required drivers for a
given design need be purchased. From an electrical standpoint, the drivers produce large pulse currents that tend to
disturb the analog circuitry close by, particularly within the
controller. By moving the drivers off chip, these pulse currents can be localized to the drivers themselves. PCB layout
is also simplified since the drivers will not need long, hi di/dt
traces. The drivers can be located very close to their respective MOSFETs. This is especially advantageous in a multi–phase design that, by it’s nature, occupies a fair amount of
board real estate. Shorter gate drive runs will also help
minimize radiated emissions from the power supply. The
result is much better-behaved control circuitry and less likelihood of needing several PCB iterations to optimize the
circuit’s performance.
UNDER VOLTAGE LOCK OUT (UVLO)
The 5V supply input has a UVLO function with hysteresis,
assuring stable, predictable start-up performance.
POWER GOOD FUNCTION
The PWRGD window is -12% to +230mV (typical) of the
programmed output voltage. The PWRGD function is
masked during VID transitions to prevent false power fail
indications. Masking time is guaranteed to be at least
100usec over the full temperature range.
INTEL SpeedStep™ TECHNOLOGY
The LM27262 supports IST. See also respective Intel specs.
IST or Geyserville-III operation is a real-time dynamic switching of the CPU core voltage and frequency between multiple
performance modes.
CURRENT BALANCING CIRCUIT
In order to ensure current balance between phases, the
LM27262 measures the instantaneous load current for the
“on” phase and forces this current to be equal to the average
of all the active phase currents.
Refer to Figure 3. Only two phases are shown for simplicity.
The circuitry in Figure 3 is duplicated on the other two
phases of a 4-phase design. The VIL pins connect to the
output side of the current sense resistors, while the VIH pins
connect to the inductor side of the sense resistors. All of the
VIL signals are summed through the internal 10kΩ resistors
so that the difference between the signals VILAVG and
VIHAVG represents the average value of all the corresponding sense voltages.
DAC ACCURACY and VREF SELECTION
The LM27262’s internal voltage reference is nominally
1.235V. Accuracy is ± 0.9% or better at room temperature.
Due to the required precision of the VRD-10 specification,
the LM27262 was designed with the ability to use an external
precision reference. Since National Semiconductor’s precision 0.2% accurate voltage references have a 1.225V typical
output voltage, the LM27262 has a 10mV internal offset
switched in when REFINT selects an external reference.
This allows compensating for the 10mV difference between
the internal and external references. The LM27201 is the
recommended external reference for use with the LM27262.
20083423
FIGURE 3. Current Balancing Circuit
Amplifier A1 converts the difference, VIHAVG-VILAVG, to a
ground-referenced signal, which represents the instantaneous average current per phase. Amplifier A2 converts the
instantaneous current information for phase A from a differential to a ground referenced signal. Amplifier A4 acts as an
www.national.com
14
STANDARD VID CODE OFFSET
Intel’s VRD-10 specification requires a “Standard Offset”.
This offset is typically 25mV but is subject to change with
future specification revisions. As such, the LM27262 has an
externally programmable offset. A resistor from the IREF pin
to ground programs a precision current thru a resistor connected between VPROG and VSTDOS pins. The recommended nominal current value is 80µA. The IREF pin forces
1.4V across the current programming resistor. The IREF
programming resistor value is therefore: R = 1.4V / 80µA =
17.4kΩ.
The VPROG output is a buffered version of the internal DAC
output. The voltage drop between this pin and the VSTDOS
pin is equal to the IREF current times the value of the offset
resistor. For a 25mV offset and RIREF equal to 17.4kΩ, the
offset resistor should be 309Ω. The error from using standard 1% resistor values is as follows: The source current can
LM27262
Operation Descriptions
(Continued)
be recalculated as 1.4V / 1.74kΩ = 80.46mA. Using a standard 309Ω, 1% value for the offset resistor produces a
nominal offset voltage of 24.86mV for an error of 0.14mV.
LOAD LINE SLOPE
The Load Line Slope (LLS) is commonly known as Adaptive
Voltage Positioning (AVP). In the LM27262 the AVP is implemented as “active voltage positioning”. Active voltage positioning synthesizes the load line actively so as to limit power
dissipation. In a typical four-phase application using 2 mΩ
sense resistors, the effective impedance due to the resistors
is only 0.5 mΩ. Yet with active voltage positioning an effective 1.3 mΩ impedance is synthesized with no additional
losses. If implemented in a purely passive manner, nearly
three times the losses would be incurred.
For an LM27262 application the LLS can be calculated as
follows:
Slope = 3.818 x RSENSE x R2/(R7+R2); where 3.818 is a
function of the gain of the LM27262’s internal load line
circuit; Slope and RSENSE are measured in mΩ. Referring
to the typical application circuit, the total resistance of the
R7+R2 resistor divider should be about 5.5kΩ. For a slope of
–1.3 mΩ and 2 mΩ current sense resistors the 1% standard
resistor values calculate as R7=4.75kΩ and R2=976Ω. The
LM27262 will automatically compensate if the number of
active phases is reduced to either two or three.
20083424
FIGURE 4. Current Sense Filtering
Total resistance of R1+R2 resistor divider should be approximately 50kΩ ± 10%. Lower values will tend to overload the
VREF output while higher values may increase the OCP
threshold error due to variations in CLIMADJ pin input bias
current.
To calculate the divider values assuming a total divider resistance of 50kΩ:
VR1 = VRS/ 0.48
R1=VR1 x 50kΩ/ VREF
R2= R1 x (VREF - VR1) / VR1
A 0.1µF filter capacitor should be connected across R1 for
reducing switching noise pickup.
Careful connection to the current sense resistors is crucial
for OCP threshold accuracy. Always use Kelvin connections
to low value sense resistors in order to minimize the effects
of trace resistance. With only a couple of µΩs of sense
resistance, a few hundred µΩs of trace resistance will result
in significant measurement errors. The connections to all
sense resistors should be as close to physically identical as
possible to ensure good phase-to-phase matching.
Generally, the worst-case low limit for the OCP threshold
should be set at least 10% to 15% above the maximum
desired continuous load current. The voltage across the
current sense resistors at the onset of current limit is approximately 48% of the voltage between the CLIMADJ pin
and Vref. Keep in mind that the current limit is pulse by
pulse, so the peak inductor current needs to be calculated to
determine the actual current limit trip point.
In order to avoid noisy current sense measurements, it is
usually desirable to add small RC filters at the current sense
inputs (see Figure 4). Typical values are on the order of 1Ω
and 0.1µF. These filters will slow down the current limit
circuit’s response time a bit and increase the actual current
limit relative to the theoretically expected value. The 48%
scale factor mentioned above includes an empirical adjustment for this. It will be necessary to verify the final value
experimentally.
OUTPUT OVER-CURRENT PROTECTION (OCP) and
PROGRAMMABLE CURRENT LIMIT
The LM27262 has a genuine OCP feature based on actual
load current measurement as a voltage drop across the
current sense resistors. Unlike some other OCP techniques,
such as a short-circuit protection based on detection of an
under-voltage condition, this true current limit approach allows a system designer to use power train components that
are not significantly over designed. There is also a time
delayed latch off feature that will be discussed later to further
protect the regulator from sever overload conditions. The
LM27262 has a CLIMADJ input that allows the voltage regulator designer to set the current limit threshold via a simple
resistor divider. Refer to Figure 4 below. The current limit is
programmed for each phase. For instance, for a 44A current
limit in a 2-phase application program 22A per phase; for
44A current limit in a 4-phase application, program 11A per
phase. In the latter case, the programming resistor R1
should be smaller. The current limit threshold will change
somewhat as a function of input voltage and die temperature, reducing somewhat at higher input line voltages and
temperatures. This is due largely to changes in inductor
ripple current. Therefore, current limiting should be tested at
the highest input voltage and operating temperature likely to
be encountered in a particular application. Some empirical
adjustment of the current limit program resistors may be
necessary.
SOFT START, VIDPGD DELAY and TURN-ON TIME
The soft-start feature minimizes inrush current and prevents
output voltage overshoot. The SOFTCAP pin has an internal
current source of approximately 3.2µA that charges a programming soft-start/soft-stop capacitor. There is an approximately 2msec built-in delay between the time that VRON is
asserted and the SOFTCAP starts charging. This allows the
VID code to settle before the switching regulator turns on.
The soft-start ramp time can be calculated using the following formula:
15
www.national.com
LM27262
Operation Descriptions
LOOP COMPENSATION
(Continued)
An RC network connected between the VFB and VCOMP
pins compensates the feedback loop’s gain/phase characteristics. These two pins are respectively, the input and
output of the error amplifier. Feedback loops such as these
are best compensated through the use of an empirical approach. The best approach is to measure the control to
output transfer function and then design an appropriate error
amplifier compensation.
TSOFTSTART-RAMP = (VCORE/ISOFTCAP)CSOFTCAP;
TVIDPGD = TSOFTSTART-RAMP x 0.5V/VCORE;
TTURN-ON = TVIDPGD + (VCORE/ISOFTCAP)CSOFTCAP; where
ISOFTCAP = 3.2µA and Volt, Amp and Farad units are used.
For example:
If TSOFTSTART-RAMP ) 5 msec for VCORE = 1.55V and Css =
10nF then TVIDPGD ) 1.6 msec and TTURN-ON = 6.6 msec. If
TSOFTSTART-RAMP ) 3.6 msec for VCORE = 1.15V and Css =
10nF then TVIDPGD ) 1.6 msec and TTURN-ON = 5.2 msec
Component Selection
SOFT STOP
POWER PATH COMPONENT SELECTION
The soft-stop feature forces a well-controlled power off transition. The output voltage ramps down smoothly, eliminating
the possibility of a large negative voltage at the output. This
feature eliminates the large need for a Schottky protection
diode or a clamp transistor at the load.
The choice of power path components is critical to achieving
a properly behaved regulator. Design considerations usually
include such things as efficiency, transient response, output
ripple, size and cost. The process tends to be somewhat
iterative while converging on a workable design.
The LM27262 has an internal 50kΩ resistor connected to the
SOFTCAP pin that discharges the SOFTCAP capacitor. The
soft-stop ramp-down time is approx. 9msec with a 33nF
capacitor, or approximately 5 x RC, where R is 50kΩ, and C
is the soft-start capacitor.
The first decision that must be made is the number of phases
to use. The maximum load current that the design must
deliver usually dictates this. With the power devices available at the time of this writing, the practical upper limit is
about 20 to 25 amps per phase. Trying to run higher per
phase load currents results in thermal problems as well as
the inability to maintain an all surface-mount design. In some
instances, it is possible to pull higher phase currents if the
peak’s duration is relatively short and the average current is
well below peak. Another possible criteria for selecting the
number of phases is to capitalize on the ripple current cancellation effects of multiphase designs. In theory, at a VIN to
VOUT ratio equal to the number of phases, the input and
output ripple currents approach zero. If the design will run
close to this “sweet spot” it may influence the number of
phases selected. For instance, adding a phase may prove
most advantageous.
One’s initial reaction to increased phase count is that the
solution becomes much more costly. But this assumption
isn’t always correct. In theory, the total energy stored in the
output inductors decreases as phases are added. This is
due in part to the ripple cancellation effects and in part to the
energy storage being a function of the square of the inductor
currents. So for equal inductor values in a two-phase design,
the energy stored is only 50% that of a single phase design.
In practice, for equal output ripple, the inductance in each
phase of a 2-phase design could be about 1⁄2 that of a
comparable single phase design. Therefore, energy storage
per inductor is only 25% that of a one phase design. So,
although there may be two inductors in the 2-phase design,
they are each much smaller, lighter and hopefully lower cost,
than the comparable single-phase solution. As for MOSFET
selection, since the total current being switched is the same
regardless the number of phases, in theory, the total Rds(on)
required is the same as well. It just gets split into more
packages in the multi-phase design. Some difficult to characterize advantages of a higher phase count relate to MOSFET parasitics. For instance, the body diode reverse recovery effects of the low side switch adversely effect the
switching loses in a buck regulator. Larger FETs for both the
low side and high side switches will have much greater
losses than smaller devices switching lower currents. Spurious turn-on of the low side FET due to its Miller capacitance
is also less problematic in smaller devices. The result is that
in many cases, the higher phase count design will prove to
be somewhat more efficient than a lower phase count design
that can provide comparable full load current.
VID-CODE CONTROLLED VCORE TRANSITIONS
The VID transition slew rate is set by an external resistor
connected between the VIDSLEW and SOFTCAP pins. This
permits an additional level of slew rate control beyond that
provided by the soft-start function.
UVP, OVP and OCP SHUTDOWN PROGRAMMABLE
Delay
If PWRGD is de-asserted for any reason, the voltage regulator can disable its output and latch itself off. Different
systems can tolerate various fault conditions for different
time durations. A programmable delay feature enables the
system designer to chose how long the supply will wait
following the detection of an OVP or UVP event prior to
shutting down. By adding a capacitor to the DELAY pin, pin
34, the latching of fault events can be delayed. If the DELAY
pin is grounded, latch off is defeated entirely. The following
formula should be used for calculating a programming capacitor value:
CDELAY = TDELAY x 12.5µA/1.4V or
TDELAY/112kΩ where C is in Farads, 1.4V is the
“DELAY Threshold Voltage”, and 12.5µA is the
“DELAY Charge Current”. For example,
CDELAY = 0.22µF programs a 25ms delay.
Grounding the DELAY pin will disable the latch off function.
This can be most helpful during system de-bug or if the
latch-off feature is not desired for some reason.
2-, 3- or 4-PHASE OPERATION
2-, 3- or 4-phase operation is user selectable. For lower
current designs it may be desirable to use fewer than 4
phases.
LOGIC INPUTS and OUTPUTS - GENERAL
All logic control inputs have hysteresis that increases noise
immunity and, particularly for the VRON signal, enables a
designer to turn the LM27262 on from a 3.3V rail via an
external RC-delay circuit. Note that the logic outputs are not
short circuit protected and must not be short-circuited to
either power rails or ground.
www.national.com
16
In general, it will be necessary to add high frequency decoupling as well as the bulk capacitance calculated above. An
array of at least 20, 22µF, 1206 case ceramics is recommended. They should be as close to the CPU as possible.
With the output capacitors chosen, an upper bound can be
established for the inductor value:
(Continued)
Since VRD-10 designs must support large load transients
while maintaining very tight output regulation, a good place
to start the design is the output capacitors.
OUTPUT CAPACITOR SELECTION
For designs that will be subjected to large load current
transients, the output capacitor array is probably the best
place to start. It is assumed that the full amplitude of the load
current step will be drawn from the output capacitors for a
short time. As such, there will be a significant droop in the
output voltage that’s a function of the step size and the
output capacitor’s impedance. The output voltage step will
have three basic components. The first is a more or less
vertical edge equal to the ESR (Equivalent Series Resistance) of the output caps multiplied by the load step amplitude or ∆I x ESR. There’s also a component equal to the ESL
(Equivalent Series Inductance) multiplied by the rate of
change of the load current, (∆I/∆t) x ESL. The ESL induced
spike is usually small in value and short lived, assuming a
clean board layout with good high frequency decoupling, and
can usually be ignored. In sizing the output capacitors, a
good starting point is to assume that the ESR step will be
20% to 50% of the allotted transient voltage spec. The low
end of the range will apply to ceramic capacitors and the
high end of the range to tantalum or aluminum electrolytic
devices. The remainder of the tolerance can be allocated to
the output capacitor’s droop voltage. The droop rate, ∆V/∆t,
is equal to ISTEP/COUT, where Istep is the amplitude of the
load transient. The total droop amplitude is equal to ∆V/∆t
multiplied by the time it takes for the regulator to get the
output voltage slewing in the opposite direction. See Figure
5 for details.
20083426
FIGURE 6. Normalized Pk-Pk Output Ripple As A
Function Of Duty Factor and Number Of Phases
L < COUT x (VIN (MIN) - VOUT(MAX)) x ESR / ∆IOUT
This value inductor should be installed in each phase. Larger
inductor values will result in a delay in the output voltage
recovery to a load step. Smaller values will store less energy
(lower cost) but will increase the output ripple. Since the
peak switch currents will also be higher, the efficiency is
likely to suffer somewhat with smaller inductors.
Assuming a minimum input voltage of 12V and 1.5V out with
a 50A load step and the capacitors selected above,
L < 2340µF x (12V- 1.5V) x 0.833 mΩ/50A
L < 0.41 µH
Something around 0.5µH will be the closest standard value
and should prove adequate. Since this value is slightly
greater than desired, dynamic performance will suffer
slightly.
If this value will yield excessive ripple current at maximum
input voltage (greater than about 40% of the single phase
DC current), then a larger inductor should be considered and
therefore, optimal dynamic performance will not be obtained.
The tradeoff is typically efficiency vs. dynamic performance.
During a load-off transition, the extra energy stored in the
inductors will end up in the output capacitors. This magnetic
energy, LI2/2, will be stored in the output capacitors as
CV2/2. The energy already in the output capacitor prior to the
transient, and that left in the inductor after the event, must
also be accounted for.
Therefore:
VMAX = [(n x L/C) x ((IMAX/n)2–(IMIN/n)2) + Vinit2]1/2
Where VMAX is the peak output voltage, n is the number of
phases, IMAX is the high load current , IMIN is the low load
current, C is the output capacitance, L is the per phase
inductor value, and Vinit is the output voltage prior to the load
dump.
From our example assuming a 70A max load and a 50A step:
VMAX = (4 x 0.50µH x ((70A / 4)2 – 20 / 4)2) / 2340µF +
1.4452)1/2
20083425
FIGURE 5. Output Transient Response
In a design with voltage positioning, the ideal ESR of the
output capacitor array should be less than or equal to the
load line slope. So for a VRD-10 design we should assume
1.5mΩ for the output capacitor ESR.
In a four-phase design, it’s likely that the latency prior to
getting a high-side switch turned on is approximately 1/4 of a
full cycle. An estimate of about twice that, or around 1.5µs, is
a good place to start for making the droop calculation. As an
example, assume a 50 amp load step and a tolerance of
85mV with high performance polymer capacitors: Using a
390µF, 5mΩ capacitor, the design requires a minimum of 4 in
parallel to meet the ESR estimate. The droop in 1.5µs would
be:
Droop = 1.5µs x 50A/1560µF = 48mV
Add this to the 75mV ESR droop and we can see the spec is
not met. Therefore several additional capacitors must be
added. Rerunning the numbers with 6 capacitors we get:
Droop = 1.5µs x 50A/2340µF = 32mV
Plus an ESR step of 50A x 0.833mΩ + 32mV = 73.6mV
17
www.national.com
LM27262
Component Selection
LM27262
Component Selection
MOSFET SELECTION
(Continued)
The choice of power FETs is driven primarily by efficiency or
thermal considerations. There are two main loss components to consider, conduction losses and switching losses.
The switching losses are primarily due to parasitics in the
FETs and are very hard to estimate with any degree of
accuracy. The conduction losses are much easier to characterize. The switching losses in the low-side FET are very low
since it’s essentially a zero-voltage switched device. However, the high-side device’s switching losses are usually
larger than its conduction losses. The primary contributor to
high-side FET switching losses is related to the reverse
recovery characteristics of the low-side FET’s body diode.
During the small dead band where both FETs are off every
cycle, the low-side FET’s body diode will carry the inductor
current. The problem is that the body diode exhibits a significant reverse recovery time, trr. During this time, the FET
looks like a short circuit. When the high-side FET is subsequently turned on, there is a shoot through path from the
input supply to ground. A larger high-side FET will tend to
exhibit a larger shoot through current. Therefore, it is undesirable to oversize the high-side device. Since the low-side
device looks like a short, the entire supply voltage is impressed across the high–side device, along with a simultaneous high current. The result is very high momentary power
dissipation. The total power lost is a direct function of the
switching frequency.
As a starting point, assume that about 1⁄2 of the switcher’s
total losses will take place in the MOSFETs. If from our prior
example, we assume a desired 90% efficiency at full load,
the FET losses are therefore 5% of the full output power or
5.25W. Since we have four phases, that works out to 1.31W/
phase. This is divided between the upper and lower FETs.
Since the step down ratio is large, the low side FETs will be
on for most of the period. The low-side FET conduction loss
is I2 x Rds(on) * (1-DF), where DF is the duty factor, VOUT/
VIN. The worst-case dissipation occurs at high input line
voltage. We’ve assumed 12V for our example. Allowing 50%
of the total FET loss to the low-side switch gives us a
dissipation of 0.66W x Rds(on) . This will usually result in a
conservative design. Solving for Rds(on) :
Rds(on) = Pdis/(I2 x (1- VOUT/VIN)) Rds(on) = 0.65W/(17 x
5A2x (1-1.5V/12V))
Rds(on) = 2.4mΩ
A pair of Si7356 MOSFETs in parallel is very close to this.
These are trench devices with excellent thermal characteristics as well.
The high-side FET on-resistance is calculated similarly, but
allow one half of the dissipation to switching losses. Therefore the allowable conduction loss is approximately 0.325W.
Rds(on) = Pdis/(I2 x (VOUT/VIN))
Rds(on) = 0.325W/(17x 5A2 x 1.5/12V))
VMAX = 1.526V
There will also be an initial step equal to ∆IOUT x ESR, which
in our example will be approximately 41mV. The two effects
are not entirely additive since the voltage across the ESR is
decreasing as the capacitor gets charged. Therefore, the
actual peak voltage will be somewhat less than the 1.567V
that simple addition would predict. In general, it’s a good
idea to provide pads for a couple of extra capacitors on the
layout in case a little extra decoupling proves necessary.
The output ripple currents of multiphase regulators tend to
cancel to some degree. This greatly reduces the demand on
the output capacitors. Figure 6 allows a simple estimate to
be made of the total output ripple current based on the
number of phases and the nominal duty cycle. Simply pick
the worst case (highest ripple) operating point off the curve
and multiply by the single channel pk-pk ripple current. The
expected pk-pk output ripple voltage will be approximately:
Vrip = Irip x ESR
This simplified equation ignores the reactive term of the
capacitor’s impedance. With any kind of electrolytic capacitor it’s generally safe to ignore the capacitive reactance term
since it will prove to be negligible compared to the ESR.
CALCULATING THE INDUCTOR RIPPLE CURRENT
The LM27262 operates at a switching frequency of 300kHz
per phase. The high side switch on time is therefore the
period, 3.33µs, multiplied the duty factor, VOUT/VIN. During
this time the inductor is connected between the input and the
output, so inductor current ramps positive during this time.
The peak-peak ripple current ∆I is approximately equal to:
∆I = 3.3µs x (VOUT - VOUT2 / VIN) / L
Continuing our example and assuming a maximum input of
12V:
∆I = 3.3µs x (1.5V - 1.5V2 / 12V) / 0.50µH
∆I = 8.66A
IIN = 1.5A
With a maximum phase current of 17.5A this is a bit higher
than desired so a little larger inductor value may be in order.
Assuming a 35% ripple current and 17.5A/phase:
L = 3.3µs x (1.5V - 1.5V2 / 12V) / 6.125
L = 0.71µH
If the larger value of inductor is used, it may be necessary to
go back and recalculate some of the early assumptions.
The peak current seen by the inductors will be the maximum
DC current plus one half of the ripple current. The maximum
DC current should be assumed to be approximately 10% to
15% greater than the maximum anticipated load current to
allow for short circuit current. The inductors must not hard
saturate in a fault. Again from our example, 110% of 17.5A
plus one half of the 6.125A peak-peak ripple current yields a
peak inductor current of 22.3A. There will usually be two
current ratings associated with an inductor. One is the average current rating and the second is the saturation current.
Only the saturation current need be considered for short
circuit limiting. The sustained DC current is the 110% of
17.5A or roughly 19A in this example. If over current latch off
is employed, only the 17.5A steady state current need be
considered since the inductor’s steady state current rating is
basically a thermal limitation.
www.national.com
Rds(on) = 8.5mΩ
A pair of Si7392 FETs will meet this requirement.
GATE DRIVE RTEQUIREMENTS
Energy for the high side gate drives is stored in the boost
capacitors, which in turn are powered by the 5V supply. The
charge stored in each boost capacitor should be between 10
times and 20 times the Qg specified for the high-side FETs.
For the example given, the specified maximum Qg is 15nC.
There will be two of these devices in parallel so the combined gate charge is 30nC. Therefore, the charge stored on
the boost capacitor should be a minimum of 300nC.
18
indicates that more capacitors than this minimum number
are required, they should be distributed evenly across the
input voltage power plane. The plane that interconnects the
phase inputs should be as large as possible in an effort to
ensure good current sharing between the input capacitors.
Be sure and install a good quality ceramic capacitor across
each phase’s FETs for high frequency bypass.
(Continued)
Since Q = C x V and the capacitor will be charged to 5V, the
minimum capacitance required is 0.06µF. A 0.1µF would be a
good choice. In some instances, it’s desirable to add small
resistors between the bootstrap capacitors and the CBOOT
pins of the drivers. This allows the high-side FET’s turn-on to
be slowed down a bit to minimize shoot through currents
associated with the low–side FET’s body diode reverse recovery time. This technique avoids slowing the high-side
FET’s turn off transition. A value of 1.0 to 5.0 ohms is usually
adequate.
Layout Considerations
Proper PCB layout is critical to having a high current DC-DC
converter work correctly. The most important part of the
layout is the power path. Start with the large, high current
parts and lay them out in a logical power flow. Avoid using
internal layers as the primary high current paths. It’s best to
connect power devices together directly with copper on the
same layer the parts are to be mounted on. Avoid vias as the
primary conductor in the high current paths. Inner layer
copper can be used in parallel with top-side copper to good
advantage. The vias that connect the layers should be allowed to solder fill. Small vias (10mil dia. or less), should be
limited to approximately 1A, while those with internal diameters of 20mils or more may be able to handle 2A or a little
more. In general, adding more vias between layers is better
than fewer.
INPUT CAPACITOR SELECTION
The input capacitors are required to deliver the difference
between the average and instantaneous currents to the
regulator in an effort to control EMI at the input. In sensitive
applications, a small inductor (typically only a few hundred
nanoHenries) should be placed in the line between the input
source and the input capacitors. This is not usually necessary in battery powered devices due to the low impedance of
the power path and the relative insensitivity of the battery rail
to the regulator’s switching noise. The most critical specification for the input capacitors is their ripple current capability. In a multiphase regulator, there is a significant amount of
input ripple current cancellation, hence a much lower input
capacitor requirement than a comparable single-phase design. Figure 7 shows the normalized input ripple current for a
given number of phases and duty factor. The inductor ripple
current is assumed to be 30% of full load current for this
analysis. The ripple current percentage only affects the
depth of the cusps in the curves. Be sure to examine the
entire range of input voltage to determine the worst-case
(maximum) ripple current. Multiply the full load output current
by the factor obtained from Figure 7 to determine the RMS
input ripple current.
Once all of the power parts are placed and routed, the
control IC can be placed and connected. Since the LM27262
uses external drivers, there are no large pulse currents in
either VCC or the ground connection. This allows the chip
ground to be remotely connected to the load’s local ground
sense point. Keep this connection under a couple of inches
in length and be sure it’s a wide trace (0.05 in or more). Avoid
locating the controller between the power switches and the
load. This minimizes ground drops between the load and the
controller. If the controller is located on the side opposite the
CPU from the power stage, there will be essentially no DC
drop across the area of ground plane between the CPU and
the controller.
There should be a good quality ceramic bypass capacitor
placed very close to the IC’s VCC and ground pins and
connected with very short traces. All of the low level analog
signals associated with the controller should be referenced
back to the IC’s local ground connection at Pin 23. A single
point ground should be established at that pin. Run a ground
trace to such things as the Softstart cap ground, Ilim divider
and external reference if used, from the primary ground pin.
The loop compensation components should be located as
close to pins 34 and 35 as possible. Unlike a typical PWM
controller, the voltage on the softstart capacitor is the actual
reference voltage used by the error amplifier. As such, it is
imperative that the SOFTCAP pin be kept as quite as possible. Again, the best approach is a dedicated analog
ground, either in the form of a separate trace that daisy
chains to all the grounded control components, or a small
plane that connects to the main ground at the chip ground
pin only.
20083427
FIGURE 7. RMS Input Ripple as a Percentage of DC
Outpout Current vs Duty Factor and Number of Phases
There should be at least one bulk input capacitor across the
power FETs of each stage. If the ripple current calculation
19
www.national.com
LM27262
Component Selection
LM27262
Layout Considerations
(Continued)
20083428
FIGURE 8. Power Path Layout
www.national.com
20
LM27262
Physical Dimensions
inches (millimeters) unless otherwise noted
48-Lead LLP Package
Order Number LM27262LQ
NS Package Number LQA48B
48-Lead TSSOP Package
Order Number LM27262MTD
NS Package Number MTD48
21
www.national.com
LM27262 Intel CPU Core Voltage Regulator Controller for VRD10 Compatible PCs
Notes
National does not assume any responsibility for use of any circuitry described, no circuit patent licenses are implied and National reserves
the right at any time without notice to change said circuitry and specifications.
For the most current product information visit us at www.national.com.
LIFE SUPPORT POLICY
NATIONAL’S PRODUCTS ARE NOT AUTHORIZED FOR USE AS CRITICAL COMPONENTS IN LIFE SUPPORT DEVICES OR SYSTEMS
WITHOUT THE EXPRESS WRITTEN APPROVAL OF THE PRESIDENT AND GENERAL COUNSEL OF NATIONAL SEMICONDUCTOR
CORPORATION. As used herein:
1. Life support devices or systems are devices or systems
which, (a) are intended for surgical implant into the body, or
(b) support or sustain life, and whose failure to perform when
properly used in accordance with instructions for use
provided in the labeling, can be reasonably expected to result
in a significant injury to the user.
2. A critical component is any component of a life support
device or system whose failure to perform can be reasonably
expected to cause the failure of the life support device or
system, or to affect its safety or effectiveness.
BANNED SUBSTANCE COMPLIANCE
National Semiconductor manufactures products and uses packing materials that meet the provisions of the Customer Products
Stewardship Specification (CSP-9-111C2) and the Banned Substances and Materials of Interest Specification (CSP-9-111S2) and contain
no ‘‘Banned Substances’’ as defined in CSP-9-111S2.
National Semiconductor
Americas Customer
Support Center
Email: [email protected]
Tel: 1-800-272-9959
www.national.com
National Semiconductor
Europe Customer Support Center
Fax: +49 (0) 180-530 85 86
Email: [email protected]
Deutsch Tel: +49 (0) 69 9508 6208
English Tel: +44 (0) 870 24 0 2171
Français Tel: +33 (0) 1 41 91 8790
National Semiconductor
Asia Pacific Customer
Support Center
Email: [email protected]
National Semiconductor
Japan Customer Support Center
Fax: 81-3-5639-7507
Email: [email protected]
Tel: 81-3-5639-7560