You are on page 1of 7

Compensator Design for Adaptive Voltage

Positioning (AVP) for Multiphase VRMs


Martin Lee*, Dan Chen*, Chih-Wen Liu*, Kevin Huang†, Eddie Tseng† and Ben Tai†
*
Department of Electrical Engineering †
RichTek Technology Corp.
National Taiwan University
Chupei City, Hsinchu, Taiwan
Taipei, Taiwan

Abstract—In an AVP scheme, feedback compensation for a variety of circumstances. Simulations will be used to
design is crucial to achieve a desirable constant verify the results. Practical conclusions will be drawn
converter output impedance while maintaining from the investigation.
converter stability. The model proposed and the Zoi Zoc
analysis given provides insightful view of the S.R. Buck VRM
iL
interaction of the two performances. Depending on Power Stage
the relative location of the noise-suppressing Vg Ro
d vo
compensation pole with respect to the capacitor Ri1
ESR-zero frequency, the performance of output
Zf vdroop
impedance and stability margin can be vary a great
deal. If the two frequencies are too close to each other, vc Z2
the stability margins get small, and Zoc exhibits a EA
Fm
spike /dip at the ESR-zero frequency. This happens in VP
a realistic case, when ceramics capacitors are used Vref
and switching frequency is set high. This conclusion is
verified in a SIMPLIS simulation of a practical Fig. 1. VRM buck Controller with AVP control.
4-phase VRM.

I. INTRODUCTION II. CONTROL BLOCK DIAGRAM


Adaptive Voltage Positioning (AVP) scheme has been
A. Compensator transfer functions
used in multiphase synchronous rectifier buck converter
topology for satisfying the power requirement of Intel’s From Fig.1, the diagram around the error amplifier EA
CPUs [1]. Use of such a scheme normally leads to lower is isolated and redrawn in Fig. 2. From this figure, one
power dissipation in the CPU and smaller output can derive that:
capacitors required in the power converter circuit. Zf
It has been pointed out that to achieve AVP the
Z2
(
vc − ⋅ vo + vdroop − vdroop ) (1)
closed-loop small-signal output impedance of the
converter circuit should be a constant value which is
specified by Intel’s requirements [2-3]. The goal of the From this equation, the equivalent circuit of the error
feedback design is therefore to achieve a desirable output amplifier compensation circuit is shown in Fig.3. The
impedance characteristic while still maintaining feedback voltage droop caused by changing the Vref voltage has
stability margin. been represented by adding the two Vdroop inputs in the
Figure 1 shows a simplified diagram of a AVP control circuit shown in Fig.3
scheme to achieve active droop control. It has been . .
pointed out that by proper switching-frequency and
component scalings, the control behavior of a multiphase
converter can be represented by a simplified single-phase
control circuit shown in Fig.1 [4]. In the diagram, the
sensed output inductor current information is subtracted
from the reference voltage and fed into the non-inverting
terminal of the error amplifier EA. And the output voltage
signal is fed into the inverting terminal. This scheme
allows the converter characteristic of output voltage
declining with increasing load current as AVP requires.
In this paper, the small-signal average model of buck
converter circuit will be reviewed first. A model will then
be developed for the AVP control scheme shown in Fig. 1.
Investigation will be conducted to the compensator design Fig. 2. Compensator circuit diagram.
Duty cycle to output voltage:
s
1+
vo ( s) ωesr
Gvd ( s) Vg ⋅
d ( s) Δ
Input voltage to inductor current:

iL( s) s⋅ Co
Gig ( s) D⋅
vg ( s) Δ
Fig. 3. The equivalent circuit diagram of that shown in Fig.2.
Output current to output voltage:
B. Control Block Diagram
The control scheme shown in Fig.1 can be represented
⎛ 1 + s ⎞ ⋅⎛ 1 + s ⎞
vo ( s) ⎜ ωesr ⎟ ⎜ ωL ⎟
by the small-signal block diagram shown in Fig. 4. In the
Zo ( s) Resl ⋅
⎝ ⎠ ⎝ ⎠
lower-right corner of Fig. 4, Vdroop signal is fed into io ( s) Δ
Sumers A and B. This is to incorporate the equivalent
circuit shown in Fig. 3 into the block diagram. Gcom(s) is Voltage loop gain:
a transfer function of the compensator and is equal to –Zf
/Z1 according to Fig. 3. With the exception of the Tv( s) ≡ Fm⋅ Gvd ( s) ⋅ Gcom( s)
lower-right corner, the block diagram is similar to that of
a conventional current-mode control buck converter. All Current loop gain:
the transfer functions have been derived before [5], and
are listed in Table I. (
Ti( s) ≡ Fm⋅ Gid ( s) ⋅ Ri1 ⋅ 1 + Gcom( s) )
Compensator gain:
−Zf( s)
Gcom( s)
Z2 ( s)
Modulator gain:
d ( s) 1
Fm
vc( s) VP
Where VP is the peak value of the saw-tooth waveform.
Parameters
Ro 1
ωo ωR
(
Lo ⋅Co ⋅ Ro + Co ) Ro ⋅Co

1 Resl
ω esr ωL
Fig. 4. Small-signal control block diagram of the AVP circuit shown in Resr ⋅Co Lo
Fig. 1
1 1 2
Q ⋅ s s
TABLE I THE TRANSFER FUNCTIONS OF THE CONTROL BLOCK DIAGRAM ωo Lo Δ 1+ +
OF FIG. 4. Resr⋅Co + Q⋅ ωo ωo
Transfer Functions Ro
Output current to inductor current: Resr : output capacitor ESR.
1+
s Ro : load resistance.
iL( s) ωesr Resl : output inductor resistance.
Gii ( s)
io ( s) Δ Zoi : output impedance with Ti closed and Tv opened.
Duty cycle to inductor current: Zoc : output impedance with both Ti and Tv closed.
s
1+ fzc : frequency of compensation zero.
iL( s) Vg ωR
Gid ( s) ⋅ fpc : frequency of switching-noise-suppressing
d ( s) Ro Δ compensation pole.
Input voltage to output voltage:
s
1+ The diagram in Fig. 4 is further simplified into that in
vo ( s) ωesr Fig. 5, where 1+ Gcom-1 is the equivalent transfer
Gvg ( s) D⋅
vg ( s) Δ function when Sumer B in Fig. 4 is eliminated.
stability and audio-susceptibility are affected. These
performances will be taken into considerations in the
discussion.
A. Desirable T2(s) to accomplish a constant output
impedance
Using (2), (3), and (5), if │Ti│>>1, and │Gcom│>>1,
T2 is approximated as follow:
s
1+
R ωesr (9)
T2 ( s ) ≅ o ⋅
Ri1 1 + s
ωR
To achieve a constant Zoc as shown in Fig. 6, T2 must
satisfy the following two conditions:
Fig. 5. Equivalent control block diagram of that shown in Fig. 4. (a) Ri1 is set to be Resr in (9).
(b) │T2│>>1 for f < fesr.
C. Useful Equations
It can be seen that T2 is independent of Gcom as long as
Loop gains:
the two assumptions, │Ti│>>1 and │Gcom│>>1, hold.
From Fig. 5, two loop gain functions, Ti and Tv, are Therefore, one can use compensation function Gcom to
defined as followed:
shape T2 and T2 stability margin without much affecting
Tv( s) ≡ Fm⋅ Gvd ( s) ⋅ Gcom( s) (2) the output impedance performance.
(
Ti( s) ≡ Fm⋅ Gid ( s) ⋅ Ri1 ⋅ 1 + Gcom( s) ) (3) The two assumptions, │Ti│>>1 and │Gcom│>>1, are
generally true for frequency much lower than fesr because
By applying the Mason’s gain formula, the two loop gain,
a compensating pole is normally placed at dc frequency.
T1(s) and T2(s), evaluated by X and Y respectively, are This will be explained later in Section III-B.
shown as below:
B. Compensation design for AVP
T1 ( s) Ti ( s) + Tv ( s) (4)
Beside output impedance, two other key issues, i.e. the
Tv ( s) (5)
stability and the audio-susceptibility, need to be taken
T2 ( s) into considerations. As described previously, T2 is
1 + Ti ( s)
unchanged by the compensation Gcom as long as │Gcom│
Converter output impedance transfer functions:
The output impedance transfer function with current and │Ti│ are both much greater than 1. Therefore, a low
loop closed and voltage loop opened is expressed as: frequency compensation pole (integrator) can be used in

Zoi ( s)
( )
Zo ( s) 1 + Ti ( s) + Fm ⋅Gii( s) ⋅Gvd ( s) ⋅Ri1
(6)
Gcom to boost the low-frequency gain of Ti and Tv and
consequently T1 so that audio-susceptibility performance
1+ Ti ( s) can be improved without much affecting T2
The output impedance transfer function with both the characteristics and the close-loop output impedance
current and the voltage loop closed is expressed as: performance Zoc. A compensation zero, fzc, should also

Zoc ( s)
( )
Zo ( s) 1 + Ti ( s) + Fm ⋅Gii( s) ⋅Gvd ( s) ⋅Ri1
(7)
be placed near the resonance frequency fo to stabilize the
1 + Ti ( s) + Tv ( s) loop gain T1. Finally, a compensation pole fpc is normally
used to suppress switching-frequency noise. Placement of
The relationships between Zoc(s) and Zoi(s) can be
fpc is flexible, depending on the amount of the noise
derived as:
attenuation required. However, if fpc is too close to fesr,
Zoi ( s)
Zoc ( s) (8) then fpc may have noticeable effects on T1 and T2
1+ T2 ( s) stability margin, and Zoc may spike near fesr. This effect
Fig. 6 shows the comparison plots of │Zo│ and │Zoi│. will be discussed in details in Sections III-E and F.
It can be seen the current loop actually increases the C. Cross-over frequewncy fesr
output impedance. Only when the voltage loop is closed
will the output impedance be brought down. If a proper From the discussion in the last section, T2 crosses over
T2 is used, the output impedance can be brought to a at fesr as shown in Fig. 7. As a result, Ti and Tv must also
constant output impedance shown by the “desirable Zoc”. crosses over at fesr. Because of (5), T2 value for
frequency beyond fesr can be decreasing with -1 slope or
III. DESIGN FOR A CONSTANT OUTPUT IMPEDANCE FOR -2 slope depending on whether a noise-suppressing pole
AVP is placed. Fig. 7 shows the plots for both cases. In either
It has been pointed out that a constant output case, Zoc is essentially the same except for frequency near
impedance is required to accomplish AVP [5]. In this fesr. │Zoc│ is more likely to exhibit a spike near f = fesr
section, discussion will be given to feedback control to for the case in Fig. 7(b). T2 stability margin will also be
accomplish such a goal. As results of accomplishing this
affected by the cross-over slope of T2.
goal, other converter performances such as converter
A quick explanation of T2 behavior is given below.
From (5), for low frequency when │Ti│>>1, T2 ≈ Tv/Ti,
for frequency when │Ti│<<1, T2 ≈ Tv. Near the cross
over frequency fesr, where │Ti│≈1, T2 behavior depends
on the phase angles of Ti and Tv at that frequency. This
affects T2 stability phase margin. And according to (8),
T2 phase margin affects Zoc behavior near fesr. Spiking of
│Zoc│ characteristics may show up near fesr because of
low T2 stability margin. This will be explained in Section Fig. 8. Vector summation of 1+Ti.
III-D.
D. Behavior of Zoc, T2, and T1 near fesr
Near fesr, the amplitude of Ti is nearly unity. However,
because of vector summation, │1+Ti│ can vary from 0 to
2 depending on ∠Ti at that frequency. Fig. 8 shows two
examples of 1+ Ti. If ∠Ti =90 degree, then │1+Ti│= 2 .
If ∠Ti =120 degree, then │1+Ti│ =1. For any frequency
at which ∠Ti >120 degree, then │1+Ti│<1 Therefore,
│T2│ function may shows a spike near the fesr according
to (5). And the ∠T2 is the difference of ∠Tv and
∠(1+Ti). By the same argument, T2 phase angle at f =
fesr affect Zoc near fesr also according to (8).
ωR ω ω o ω esr
L E. Effects of switching-noise-suppressing compensating
Fig. 6. Plots of │Zo│, │Zoi│ and │Zoc│. pole fpc on Zoc, T1 and T2.
As mentioned in Section III-B, a compensation pole fpc
Tv ( s) may be used to suppress hardware switching frequency
Ro/Resr
noise in the circuit. To be effective, fpc must be
significantly lower than the switching frequency. If fpc
>> fesr, then the use of fpc has no direct effect on T2 and
Zoc behavior. However, if fpc is close to or even less than
Ro/Resr Desirable T2 fesr, then that may significantly degrade T2 stability
margin and Zoc may exhibit a spike near fesr. This is
Ti ( s) because fpc significantly affects Ti and Tv phase angle at f
ωR ω zc ω o = fesr, which affects Zoc and T2 at that frequency
according to the discussion given in Section III-D.

ω esr F. Compensation for circuit using ceramic capacitaors


Capacitor ESR zero, fesr, for ceramic capacitors is
(a) normally much larger compared with a electrolytic
capacitors such as a commonly used OSCAN capacitors.
Tv ( s) Therefore, fesr is much closer to or even greater than
switching frequency. This means that fpc in such a case is
close to or even less than fesr. Similar to the case
mentioned in Section III-E, T2 and Zoc behavior would be
degraded. In other words, T2 stability margin would be
reduced and │Zoc│ would display a spiking behavior.
Ti ( s) Two practical examples will be shown in the next section
to illustrate the point.
ωR ω zc ω o
ω pc
IV. DESIGN EVALUATION AND SIMULATION RESULTS
ω esr
Using the model and the expressions given above, T2
(b) and Zoc functions will be plotted for two practical cases.
Fig. 7. Plots of │Ti│, │Tv│ and │T2│, In case I, large OSCAN capacitors were used in which
(a) noise-suppression pole ωpc is much larger than ωesr and
fesr is relatively small compared to switching frequency.
(b) noise-suppression pole ωpc is placed at ω=ωesr.
In case II, small ceramics capacitors were used in which
fesr is close to switching frequency.
In both cases, a noise-suppression compensation pole
fpc was used. To be effective, fpc must be significantly
below switching frequency. The exact component values
of the two cases are listed in TABLE II.
TABLE II KEY COMPONENT PARAMETERS
Circuit Parameters
CASE I. using large capacitor type
Co :Oscon, (820 μF/ 12 mΩ) x12 → 9,840 μF/ 1 mΩ,
fesr=16.17 kHz.
Lo :470 nH, per-phase
fsw :300 kHz
fpc :100 kHz
CASE II. using ceramic capacitors
Co :Ceramic, (100 μF/ 1.5 mΩ) x4 → 400 μF/ 1 mΩ,
fesr=1.1 MHz.
Lo :100 nH, per-phase 10-(a)
fsw :1 MHz
fpc :300 kHz
Lo : output inductance
Co : output capacitance
Resr : capacitance ESR

In both cases, SIMPLIS simulations were run for a


practical 4-phase VRM to verify the results [6]. The
circuit diagram for a SIMPLIS simulation is shown in Fig.
9. From Fig.10 to 14 shows the T2 and Zoc plots for the
two cases. Results obtained both the theoretical
average-model calculation and the SIMPLIS simulations
are shown. The agreement between the calculated and the
simulated is generally good. It can also be seen that
│Zoc│ exhibits a large spike /dip near f = fesr which leads
to less ideal step-load response. Fig.14 and Fig. 15 show
the load-step response for the two cases. It can be seen
that case I gives a better step response because of no
voltage overshoot at step instant.
10-(b)
Fig. 10. Theoretical T2 plots for (a) CASE I and (b) CASE II.

│Zoc│

Fig. 9. Circuit diagram of SIMPLIS simulations for a 4 phase VRM


buck converter 11-(a)
│Zoc│

11-(b)
Fig. 11. Theoretical Zoc plots for (a) CASE I and (b) CASE II.

13-(b)
Fig. 13. Simulated Zoc plots for (a) CASE I and (b) CASE II.

12-(a)

Fig. 14. Simulated step-load transient response for CASE I.

12-(b)
Fig. 12. Simulated T2 plots for (a) CASE I and (b) CASE II.

Fig. 15. Simulated step-load transient response for CASE II.

V. CONCLUSIONS
A Small-signal model was developed for a multiphase
synchronous buck converter with AVP control. Based on
the model, various compensation design issues were
explored. Depending on the relative location of the
noise-suppressing compensation pole frequency with
13-(a) respect to the capacitor ESR-zero frequency, the
performance of converter output impedance Zoc and
stability margin near the cross-over frequency may vary a
great deal. If the two frequencies are closer to each other,
the stability margin get smaller, and Zoc is more likely to
exhibit a spike/ dip at the ESR frequency and step-load
response gets overshoot. This happens in a realistic case
when ceramics capacitors are used. This conclusion is
verified in a SIMPLIS simulation of a practical 4-phase
VRM.

ACKNOWLEDGEMENT
This work was supported by grant from National
Science Council of Taiwan (R.O.C.) under Award
Number NSC 94-2213-E-002-122 to National Taiwan
University. Also the authors would like to thank Transim
Technology Corporation, U.S.A. for providing SIMPLIS
simulation tool.

REFERENCES
[1] M. Zhang, “Powering Intel Pentium 4 processors,” Intel
Technology Symposium, 2000.
[2] K. Yao, Y. Meng P. Xu and F.C. Lee, “Design considerations for
VRM transient response based on the output impedance,” in Proc.
IEEE APEC, 2002, pp. 14−20
[3] K. Yao, Y. Meng and F.C. Lee, “Control bandwidth and transient
response of buck converters,” in Proc. IEEE PESC, 2002, pp.
137−142.
[4] P.L. Wong, “Performance improvements of multi-channel
interleaving voltage regulator modules with integrated coupling
inductors,” Dissertation of Virginia Polytechnic Institute and State
University, March 2001.
[5] R.B. Ridley, B.H. Cho and F.C. Lee, “analysis and interpretation
of loop gains of multiloop-controlled switching regulators,” IEEE
Trans. Power Electron., ϖολ.3, ππ.489−498, Oct. 1988.
[6] B. Wang, S. Wang, D. Chen, K. Huang, B. Tai and E. Tseng,
“Practical Simulation of Control Characteristics of a
Current-Mode DC/DC Converter,” in Proc. IEEE PESC, 2006.

You might also like