Slup 413
Slup 413
Reproduced from
2024 Texas Instruments Power Supply Design Seminar
SEM2600
Topic 3
Matthew Schurmann and Mohamed Ahmed
Literature Number: SLUP413
Introduced in 2019, the trans-inductor voltage regulator (TLVR) topology offers major transient response, power density and solution cost
improvements (a >40% capacitor reduction for the design example reviewed in this topic) versus the traditional multiphase buck voltage
regulator topology. This topic covers the operating principles of the TLVR topology, performance and cost improvements over traditional
voltage regulators, design equations, and guidelines.
VIN
Introduction
SW1
Load transient regulation performance continues to be an
important challenge in the design of voltage regulators
for modern computing devices such as microprocessors,
graphics processors, application-specific integrated
circuits and field-programmable gate arrays. Technology
trends in the development of these computing devices,
VIN
such as rapidly increasing complexity, silicon process-
node evolution, physical limitations of transistor scaling SWn
Converter Transient Response Equation 1 shows the relationship between the total
output deviation ΔV, COUT, and the rate (slope) at which
Figure 3 shows a simple block diagram of a voltage
the converter can ramp its current up or down:
regulator system subject to a load transient condition.
ISUM represents the sum of the individual inductor 2
1 ×t 1 × Istep
currents from each phase in the converter. ILOAD ΔQ resp × I step 2 Slope
ΔV = C = 2 Cout = Cout (1)
out
represents the actual load current drawn by the load
device. Any time ILOAD changes, the voltage regulator For the traditional multiphase buck converter, this slope
responds by changing the effective duty cycle of is directly related to the output filter inductance used
switching in each phase in order for the ISUM to ramp for each phase. Reducing the inductance value would
up or down to track the new ILOAD value. indeed improve the transient response of the converter.
The output filter of the converter – in particular, the filter Simply reducing the output inductance of each phase
inductance – limits how quickly ISUM can ramp to the new has unintended consequences for the converter’s power
ILOAD value. During the time when ISUM is ramping up losses and its steady-state ripple, however. Reducing
or down, the filter capacitors must supply the difference the inductance value leads to a higher inductor current
between them over time; this is known as the charge ΔQ. ripple and consequently a higher voltage ripple on the
The output voltage of the converter will undershoot or output of the converter, which typically also has stringent
overshoot during this time, and the only ways to limit the requirements. It also increases the root-mean-square
voltage deviation (ΔV) are to either increase the rate at (RMS) current in each phase, reducing overall converter
which ISUM can ramp (by reducing the filter inductance, efficiency.
for example) or to increase the total output capacitance
In a multiphase buck converter topology, the inductance
(COUT) of the filter.
value is a constant, in both steady state and
ISUM
Voltage
ILOAD
during transient events. Therefore, the selection of
Load
Regulator
an inductance value is a balanced trade-off between
transient response, power loss, and voltage ripple and
current ripple. It is not practical to make the inductance
Figure 3. Converter load transient block diagram. very small; thus, a large amount of COUT may be required
to limit ΔV in order to meet the specifications.
Figure 4 shows the typical ISUM and output voltage
The TLVR topology addresses this problem by allowing a
waveforms in a traditional multiphase buck converter.
different effective filter inductance in different conditions.
ISUM(BUCK)
IOUT A high effective value of filter inductance during steady-
VOUT
state operation limits the converter ripple and RMS
ISTEP ΔQ ΔQ
power losses. A low effective inductance value during
transient conditions dramatically reduces the amount
of COUT required to meet a given transient regulation
specification. Figure 5 shows the typical load transient
response of a TLVR converter, having a much higher ISUM
slope during the converter response.
Time
ISUM(TLVR)
IOUT Equation 4 and Equation 5 express the effect of RLL on
VOUT
the required COUT of the converter:
2
1 × ISTEP
ΔQunder 2 Slope
ISTEP ΔQ ΔQ
COUT min, step up = ΔV = ΔV + R × I (4)
under ac LL step
2
1 Istep
ΔQ ×
2 Slope
COUT min, step down = ΔVover = ΔV + R × I (5)
over ac LL step
ISUM
IOUT
VOUT = VNOM, RLL = 0 mΩ
VOUT > VNOM, RLL ≠ 0 mΩ
Time
topology using DC load line (DCLL), also known as VOUT > VNOM
ΔVAC
Time
buck converter or TLVR topology and does not change Figure 6. DC load line, or adaptive voltage positioning.
fundamentally.
Magnetics
Given a specification, in terms of a load step size and
minimum and maximum allowable output voltage, the Because the TLVR topology achieves its transient
converter typically regulates the output voltage to a benefits by allowing different effective inductance values
constant value regardless of the load current – this is in steady-state and transient conditions, it is helpful to
known as zero load line, RLL = 0 mΩ. Then the allowed explore the behavior of the coupled inductor structure
output voltage overshoot (ΔVovershoot) and undershoot that it uses. This concept is not entirely unique to the
For a non-zero-load-line design, configure the converter structure in which the windings for individual phases in
to set its output voltage as a function of the sensed the converter share a common magnetic core. Current
load current. The voltage at zero load (V0) is configured in one winding directly induces current in the others, as
to a value near the maximum allowed output voltage. the magnetic flux in the core is additive. During a load
Equation 2 describes the output voltage when using the transient, a current change in one phase (one winding)
(K≅1.0) is not beneficial, as it increases the current ripple secondary-side current to all phases achieves coupling
of the converter in steady state. Very low coupling simply between cores (the phases), as they are connected in a
reduces the transient benefits achievable. loop.
A. B.
PRI SEC
1 2
IPRI1
IL1 IL2
LLKG
SEC
PRI ISEC
Phase
2
Phase
1
Figure 7. Traditional two-phase inverse-coupled inductor. Similar to a traditional coupled inductor, it is beneficial
to have the coupling coefficient (α) between phases in
Adoption of the traditional coupled inductor for high-
the range of 0.4 to 0.7. The secondary loop controls
phase-count designs (more than four phases) has been
this coupling. The inductance in the secondary loop may
limited for several reasons. Extending it to higher
be very low, leading to high coupling (and thus a large
phase counts requires a complex core geometry to
steady-state current ripple) or simply not well-controlled,
maintain coupling symmetry. This structure also requires
as a result of interconnect and physical construction
more customization of inductors for different designs,
tolerances.
limiting scalability; for example, you would need a
different inductor for two- and three-phase designs. To control the coupling between phases, the TLVR
Additionally, until recently, aggressive patent protection topology often uses a separate physical inductor on
limited multisourcing options; no such limitation exists for the secondary side, LC, shown in Figure 9. If the
the TLVR topology. leakage inductance in the secondary-side loop is large
enough compared to the magnetizing inductance of the
The TLVR topology relies on a similar principle but with
individual coupled inductors, and can be well-controlled
a different magnetic structure, known as an indirect-
by manufacturing, a separate physical LC is not needed,
coupled inductor, shown in Figure 8. Each phase
especially in high-frequency designs switching at higher
inductor has its own physical core with two windings,
than 1 MHz per phase.
so this structure is easily scalable to higher phase
counts simply by adding more cores. The magnetizing
inductance (LM) of each coupled inductor provides
energy storage and filtering. The K between two windings
on one core can be very high. Passing the same
PRI SEC
TLVR Topology Operating Principles
Steady-State Operation
IPRI1
Figure 11 shows a typical TLVR converter schematic,
LLKG with important nodes, voltages and currents labeled.
LC
Figure 12 illustrates the steady-state operating
ISEC waveforms of a TLVR converter, with four phases shown.
In this example, the pulses from adjacent phases do
not overlap in time. There is no maximum duty-cycle
requirement for the TLVR topology. The same principles
apply for higher-duty-cycle applications where pulses do
Figure 9. Indirect-coupled two-phase inductor with a physical overlap in time.
compensating inductor.
Figure 12 shows the voltage and current waveforms of
Figure 10 shows the typical construction of a TLVR the LC of the secondary-side loop, switch nodes of all
inductor. The inductor size and shape are similar four phases, and the primary-side current of phase 4
to traditional high-current ferrite core inductors for (IPRI4). For clarity, this figure includes labels for the three
multiphase buck converters, with the secondary winding distinct states of operation.
inside the primary winding. The land pattern on the
The most important relationships are those of the LC loop
bottom of the package enables co-layout with both TLVR
and its influence on IPRI and ISUM.
and non-TLVR designs on the same physical printed
circuit board (PCB). VIN
SW1
Lm
ISUM
VIN
B. A.
SW4
A. Primary winding
B. Secondary winding Lc
Source: Eaton
+
LC ILC
–
(VIN – VOUT)
Table 1 summarizes the state of each of the relevant
SW1 – VOUT
SW2
SW3 voltages and currents shown in Figure 12, with respect
SW4
to the derivation of IPRI4 shown in the plot.
(VIN – × VOUT)
The voltage across the LC is always equal to the sum Load Transient Step-Up
of the magnetizing voltages across all phases, as shown
Figure 13 and Figure 14 show a simulated comparison
in Equation 9. LC itself always follows the fundamental
between a multiphase buck converter and a TLVR
inductor relationship, expressed by Equation 10:
design under the same load step-up condition. Table
ΔVLC = VLm1 + VLm2 + … (9) 2 summarizes the simulation parameters. These are
closed-loop simulations using the TI TPS536C9T
ΔV
ILC = L LC (10) DCAP+™ constant on-time controller.
C
• During the transient response, the multiphase buck Parameter Description Value
VIN Input voltage 12 V
converter design required many more pulses to
VOUT Output voltage 0.8 V
respond than the TLVR design, meaning that the TLVR
Total operating phase
NTOTAL 4 phases
design delivers more energy per pulse during the number
transient event. fSW
Switching frequency per
600 kHz
phase
• Given the nature of constant-on-time control, pulses 25 A to 325 A,
ISTEP Load step size
overlapped during the transient response. The LC instantaneous
Magnetizing inductance LM
voltage increased to a level significantly higher than
LM/LBUCK for TLVR, filter inductor LBUCK 150 nH/150 nH
the input voltage during pulse overlap operation, then for buck
PWM1
Table 2. Simulation parameters for transient load step-up and
PWM2 step-down examples.
PWM3
PWM4
Following the relationships described in the Steady-
IL1 State Operation section, it is evident why the TLVR is
to
IL4 able to ramp its ISUM up more quickly than the buck
converter, and why its transient response was superior.
IOUT
ISUM ISUM for the buck converter is simply the sum of its
individual inductor currents, as shown in Equation 13.
VOUT
Large For the TLVR design, ILC gets added once for each
undershoot
(100 mV/
phase, in addition to each magnetizing current (ILM), as
div)
shown in Equation 14:
Time (1µs/div)
Figure 13. Multiphase buck converter. ISUM buck = IL1 + IL2 + … (13)
VIN − VOUT V
↑Slope buck ≅ NON L − NOFF OUT
L (16)
more quickly to transients than a traditional multiphase Figure 15. Multiphase buck converter.
buck design. Shorter PWM low period
PWM1
Load Transient Step-Down PWM2
PWM3
Figure 15 and Figure 16 show a simulated comparison PWM4
Ipri1
between a multiphase buck converter and a TLVR to
design under the same load step-down condition. This Ipri4
Small Q
tresp ≅ 3µs
simulation uses the same parameters as those in Table 2. IOUT
ISUM
VOUT
↓ Slope buck = – NTOTAL L (19)
Using a similar analysis, Equation 20 shows the falling After building up a large current, the LC current naturally
ISUM slope for the TLVR design, assuming that the TLVR decays to zero, with a relatively high time constant, τLC,
magnetizing inductance LM is equal to the buck filter as described in Equation 23, formed by the LC and
inductor L for comparison purposes. The TLVR design the resistances in the LC loop. During high-frequency
ramps down its ISUM faster given the factor from the LC repetitive transients, ILC may not settle fully but will not
loop, which decreases proportionally to the square of the saturate, as load steps up and down push ILC in different
number of phases, NTOTAL. directions. Figure 17 and Figure 18 show a simulation of
this behavior:
N × VOUT
↓Slope TLVR ≅ ↓Slope buck – NTOTAL × TOTAL
LC (20)
C L
τLc = R (23)
DCR, Lc + Ntotal × RDCR, secondary + Rrouting
LC Inductor Selection
The LC has somewhat unique requirements compared
ISUM
to other inductors in a typical DC/DC design. The IOUT
inductance of LC is a trade-off between current ripple
and transient response benefits. Typically, start with
IPRI1
LC = LM as a balanced trade-off. Values between 0.8 to
IPRI4
to 1.5 times LM are common with discrete designs.
Lower values may be more common in highly integrated
designs, such as power modules.
ILC
At steady state, LC carries no DC current – only a small
0 100 200 300 400 500 600 700 800
AC current ripple – because it is switching at a high Time (µs)
frequency (at least NTOTAL × fSW when there is no pulse
fSW < 1 kHz
overlap). Its current ripple dominates its RMS current at
Figure 17. Low-frequency transient event.
steady state, described in Equation 21. Consider low
core-loss materials, such as ferrite cores, because of
the high fSW. Another option to further improve transient
ISUM
response may be soft-saturating cores. IOUT
ΔILc
Irms Lc ≈
3
(21) IPRI1
to
IPRI4
However, LC can continue to build large amounts
of current during transient events, as expressed by
Equation 22, where tRESP is the response time of the
controller, as highlighted in Figure 15 and Figure 16. ILC
Therefore, size the LC with a high saturation current, 0 5 10 15 20 25 30 35 40
Time (µs)
similar to the coupled inductors used in each phase.
fSW = 65 kHz
NON step × VIN − NTOTAL × VOUT
ISAT Lc ≫ tRESP × (22) Figure 18. High-frequency transient event.
Lc
The voltage across the LC, ΔVLC, can exceed the input perfectly (with NTOTAL × D = 1, 2, …). However, for
voltage, VIN, during a load step response. Assuming that typical applications (highlighted in Figure 20 for typical
a controller turns on NON phases in response to the load output voltages of 1.0 V, 1.2 V and 1.8 V), TLVR designs
step, Equation 24 calculates ΔVLC: typically have a 25% to 50% larger ISUM ripple, and
consequently, a 25% to 50% larger output voltage ripple.
ΔVLC max = NON step × VIN − NTOTAL × VOUT (24)
For many cases this will not be an issue, because the
Creepage is not generally a concern, as the high voltage COUT required to meet the transient requirement is much
is not sustained for a long period of time. But the high larger than the capacitance required to meet the design’s
transient voltage across LC may be important to know ripple requirement.
for application safety and component reliability in some ZPCB ZPKG
cases.
Steady-State Ripple
ISUM
Load
TLVR-based designs tend to have larger output voltage
ripple than their multiphase buck converter counterparts. ZPCB ZPKG
12 V/ 1.0 V
12 V/ 1.2 V
12 V/ 1.8 V
TLVR
Buck
degrees/NTOTAL with respect to each other. For TLVR
Ripple Current Difference (A)
40
VIN VIN
Power Loss and Efficiency
SW2
SW1
PWM1 PWM2 Figure 23 compares the power efficiency between a
multiphase buck converter and TLVR when designed
ISUM1
with the same component values. The curves are already
ISUM2
quite similar, but the TLVR design is a small amount
VIN VIN
lower (0.1%) in terms of efficiency.
SW4
SW3 While this plot is useful for demonstration purposes,
PWM11 PWM12 typically, multiphase buck and TLVR designs will not
have the same inductance values. The buck converter
will require a lower inductance value to meet the same
ILC1 ILC2 transient specifications, which further reduces its power
efficiency. In practice, when designing two converters to
Figure 21. Interleaved TLVR design. the same specifications, the multiphase buck and TLVR
converters have approximately equivalent efficiency. In
ILC1
ILC2
some cases, TLVR designs can have slightly higher
PWM1
efficiency.
95%
TLVR It is also common to use dynamic phase shedding
Buck
94% (DPS) in high-phase-count designs to improve light-load
93%
efficiency. Switching a fewer number of phases when
the total output current is low enough to be supported
Efficiency
91%
Phases can be in one of three states: high-side MOSFET
on, low-side MOSFET off; high-side MOSFET off, low-
90%
side MOSFET on; or both MOSFETs off. Typically,
89% nonlinear control techniques add or drop phases quickly
0 50 100 150 200 250 300 350 400 450
Low-Side MOSFET Current (mA)
during load transient events, so the impact on the load
transient response is minimal. Figure 25 shows the
VIN = 12 V VOUT = 1.80 V
current flow in each state.
fSW = 600 kHz RLL = 0.5 mΩ
In a TLVR design, the LC loop continues to conduct
LM = LBUCK = 120 nH LC = 120 nH
current through the body-diode phases in the third state
Excludes PDN conduction losses (both MOSFETs off), which are not switching. There will
be additional power losses from nonswitching phases
Figure 23. Efficiency vs. output current.
caused by the voltage drop of the body diodes, Vdiode.
TLVR
Therefore, for phase shedding to make sense, the
Buck
switching losses saved by not switching a phase must
IPEAK be greater than that created by the body-diode losses.
IOUT . IOUT(avg) Equation 28 describes the power losses in nonswitching
phases:
IVALLEY
ΔI 2
IRMS LSFET = IOUT × 1−D × 1 + 13 × 2 × PP
IOUT (27)
VIN
ILC
Phase Multiplication
As power requirements continue to increase rapidly, it is
often necessary to design very high phase-count (more
VIN
than 16 phase) designs using controller devices that do
not have enough independent pulse-width modulation
(PWM) outputs to control each phase individually. It has
become common to phase double or phase multiply –
that is, to drive more than one power stage with the
VIN
90%
For all phases (doubled or not) in the same LC loop,
the secondary sides are connected in series. The current
Efficiency
88%
86%
feedback lines from each phase (not shown in Figure
84%
82%
DPS 27) can be resistor-averaged for power stages with
Enabled
Disabled
80%
0 50 100 150 200 250 300 350 400
voltage-source-output current sensing, or simply added
Output Current (A)
for power stages with current-source-output current
VIN = 12 V VOUT = 1.80 V
sensing. It is possible to connect temperature sense
fSW = 90 kHz LM = LC = 100 nH outputs from each power stage (also not shown in Figure
8 phases Dual-side layout 27) together, regardless of which LC loop the power
stages are in.
TLVR CSD08860 (90-A SPS)
VIN VIN
PWM1 PWM2
VIN VIN
PWM1 PWM2
VIN VIN
PWM11 PWM12
VIN VIN
PWM11 PWM12
Lc pad is
high voltage
( 50V+)
Ph 1 Ph 3 Ph 5 Ph 7 Ph 9 Ph 2 Ph 4 Ph 6 Ph 8 Ph 10
Lc
L1 L3 L5 L7 L9 L2 L4 L6 L8 L10
Smart power stages optimized for TLVR designs must CSD95510 90-A peak, 50-A RMS 4×6 Voltage
CSD95560 90-A peak, 50-A RMS 4×6 Current
also be rated for increasingly high RMS currents and be
CSD95520 60-A peak, 30-A RMS 4×5 Voltage
able to support peak current pulses nearly two times
CSD95570 60-A peak, 30-A RMS 4×5 Current
their RMS rating for short durations, thermally as well as
Table 3. TLVR-optimized smart power stages.
electrically.
Figure 30. Worst-case overshoot (multiphase buck converter). Figure 31. Worst-case overshoot (TLVR).
increasingly popular. In this paper, we introduced the Supplies: Advantages and Compromises.” EETimes,
example designs, and practical considerations for TLVR • Lu, Zengyi, and Wei Chen. “Multi-Phase Inductor
designers. Coupling Scheme with Balancing Winding in VRM
Applications.” Published in Proceedings of the 22nd
Additional Resources Annual IEEE Applied Power Electronics Conference
• Technical Disclosure Commons. “Fast Multi-Phase and Exposition, Feb. 25-March 1, 2007, pp. 680-684.
Trans-Inductor Voltage Regulator.” Technical • Zhu, Feiyang. “Multi-Phase Coupled Inductor Analysis
Disclosure Commons Defensive Publications Series, for Multi-Phase Voltage Regulators.” Center for Power
May 9, 2019. Electronics Systems PMC Review, June 2021.
• Radhakrishnan, Kaladhar, and Jonathan Douglas, • Jiang, Shuai, Xin Li, Mobashar Yazdani, and
“Microprocessor Power Delivery Challenges.” APEC Chee Chung. “Driving 48V Technology Innovations
2022, March 22, 2022. Forward – Hybrid Converters and Trans-Inductor
• Parisi, Carmen. “Multiphase Buck Design From Start Voltage Regulator (TLVR).” Published in 34th Annual
to Finish (Part 1).” Texas Instruments application IEEE Applied Power Electronics Conference and
report, literature No. SLVA882B, April 2021. Exposition, March 15-19, 2020.
• Erickson, Robert W., and Dragan Maksimovic. 2020.
“Fundamentals of Power Electronics, Third Edition.”
New York: Springer AG.
Mailing Address: Texas Instruments, Post Office Box 655303, Dallas, Texas 75265
Copyright © 2024, Texas Instruments Incorporated