by Wing-Hung Ki
Integrated Power Electronics Laboratory
ECE Dept., HKUST
Clear Water Bay, Hong Kong
www.ee.ust.hk/~eeki
International Symposium on Integrated Circuits
Singapore, Dec. 14, 2009
1. IC Design of
Power Management Circuits (I)
Wing-Hung Ki
Integrated Power Electronics Laboratory
ECE Dept., HKUST
Clear Water Bay, Hong Kong
www.ee.ust.hk/~eeki
International Symposium on Integrated Circuits
Singapore, Dec. 14, 2009
2. Tutorial Content
1. Switching Converters: Fundamentals and Control
2. Switching Converters: IC Design
3. Switching Converters: Stability and Compensation
4. Fundamentals of Bandgap References
5. Development of Integrated Charge Pumps
6. Introduction to Low Dropout Regulators
Ki
2
4. Content
Steady State Analysis
Lossless elements
Buck, boost, buck-boost power stages
Volt-second balance
Continuous conduction mode
Discontinuous conduction mode
Ringing suppression
Pseudo-continuous conduction mode
Efficiency
Performance Evaluation Parameters
Control Topologies
PWM voltage mode control
PWM current mode control
Ki
Single-Inductor Multi-Input Multi-Output Converters
4
5. Linear Regulator has Low Efficiency
Idd
MN
IQ1
IQ2
VREF
Vdd
EA
Vo
IQ3
Io
R1
bVo
R2
C
RL
power converter
Efficiency of linear regulator is not high:
η=
Ki
Po
VI
V
Io
V
= o o = o
< o <1
Pin VddIdd Vdd Io + IQ Vdd
Can one design a power converter with efficiency close to 1?
5
6. Switches as Lossless Components
A power converter with high efficiency needs lossless components.
Reactive elements: capacitors, inductors
Active elements: switches
L
C
−
store &
relax
store &
relax
PC = 0
Ki
+
Vsw
Isw
PL = 0
+
Vsw
Isw
−
switch open
switch closed
Psw = Vsw×Isw
= Vsw×0
=0
Psw = Vsw×Isw
= 0×Isw
=0
6
7. Switching Converter: Heuristic Development (1)
Vo = Vdd
Vo
Vdd
RL
t
No regulation
SW1
Vdd
Vo
Vdd
Vo = DVdd
RL
t
duty ratio = D
Ki
Load cannot accept a
pulsating supply voltage
7
8. Switching Converter: Heuristic Development (2)
Vo
L
SW1
Vdd
C
Vx
Vdd
Ki
Vo
L
SW1
SW2
RL
C
Add a lossless filter to
achieve small ripple voltage,
but …
when switch is off, inductor
current cannot change
instantaneously and cause
spark (volt-second balance).
Vdd
Vo = DVdd
RL
Add a second switch that operates complementarily to arrive at
a functional switching converter.
t
8
9. Buck, Boost and Buck-Boost Converters (1)
Vx
SW1
Vdd
SW2
Vo
L
C
RL
Buck
Vx
L
Vdd
Vo
SW2
SW1
C
SW1
Vdd
Ki
Vo
SW2
L
C has to be in parallel with RL for
filtering, leaving three ways to place
L, SW1 and SW2 between Vdd and RL.
RL
Boost
Vx
One L and one C gives a second
order switching converter.
Three types of converters:
Step-down: buck
Step-up: boost
Step-up/down: buck-boost
(Boost-buck, or Cuk, is a 4th order
converter)
Buck-boost
C
RL
9
10. Buck, Boost and Buck-Boost Converters (2)
state 1
Vx
MN
Vdd
L
i
D1
C
state 2
RL
L
Vx
i
MN
state 1
Vo
D1
C
RL
Boost
MN
Vdd
Ki
state 1
SW1 is the controlling switch that
determines the duty ratio D, while SW2
provides a path for the inductor
current i to flow when SW1 is off.
Buck
state 2
Vdd
Vo
Vx
D1
Vo
i
L
state 2
C
RL
Buck-Boost
SW1 can be a power NMOS (MN). If
power PMOS is used, the phase has to
be reversed.
To prevent i from going negative, SW2
is usually implemented by a diode
(D1), but the forward drop gives a low
efficiency.
Note that Vo of buck-boost is negative.
10
11. I-V Relations of C and L
The I-V characteristics of a capacitor and an inductor are described by
ic = C
dv c
dt
ic
C
+
vc
i
v =L
di
dt
+
L
v
−
−
Approximations are very useful in many calculations:
ic = C
ΔVc
Δt
v =L
Δi
Δt
For sinusoidal steady state, the phasor relations are:
zc =
Ki
vc
1
=
ic
jωC
z =
v
= jω L
i
11
12. Volt-Second Balance
Switching actions cause ripples for both inductor current (i ) and
capacitor voltage (vc). In the steady state, both quantities return
to the same value after one cycle.
i
+ v
V (S1 )
= m1
L
−
di
dt
⇒ ΔI =
i
ΔI
I
L
v =L
V (S2 )
= −m2
L
V
Δt
L
0A
t1
t2
(or DT) (or D ' T)
Inductor current has to obey volt-second balance (VS balance):
V (S1)×t1 + V (S2)×t2 = 0
⇒
m1t1 = m2t2
or
m1D = m2D’
It is used to compute the conversion ratio M = Vo/Vdd.
Ki
12
13. Inductor, Input, Switch, Diode and Tail Currents
i
Consider the buck converter:
idd
is
L
Vx
MN
Vdd
i
D1
id
it
Vo
ic
Io
C
RL
Input current idd: current through Vdd
idd
is
Switch current is: i in State 1
Diode current id: i in State 2; even if diode
is implemented by NMOS switch
Tail current it: current through the
combination of C and RL.
id
Capacitor current ic: ac part of tail current
Load current io: averaged tail current
Ki
it
Io
13
14. Continuous Conduction Mode
The converter is operating in continuous conduction mode (CCM)
if the inductor current is always larger than zero.
Boost converter
(Step-up)
Buck converter
(Step-down)
Vdd
Vo
+V −
S1
S2
m1D = m2D’
⇒ (Vdd-Vo)D = VoD’
⇒ M=
Ki
V0
=D
Vdd
Vdd
Buck-boost converter
(Step-up/down)
Vo
+V −
S1
S2
m1D = m2D’
⇒ VddD = (Vo-Vdd)D’
⇒ M=
V0
1
=
Vdd 1 − D
Vdd
Vo
S1
+
V
−
S2
m1D = m2D’
⇒ VddD = -VoD’
⇒ M=
V0
−D
=
Vdd 1 − D
14
15. Discontinuous Conduction Mode
When the switching converter is operation in CCM, one switching
cycle has two states S1 and S2. When the load current becomes
smaller and smaller, eventually the inductor current would fall to
zero, and the converter then operates in discontinuous conduction
mode (DCM) with a third state S3. During D3T, all switches are open.
V (S1 )
= m1
L
V (S2 )
= −m2
L
V (S3 )
=0
L
i
ΔI
i =0
DT
D2 T
D3 T
VS balance becomes:
m1D = m2D2
Ki
15
16. Ringing Suppression
When both switches are open, L,
C and the parasitic capacitor Cx
at Vx form a resonance circuit
that leads to serious ringing.
Vx
Vdd
SW2
i
Cx
SW3
Vx
Vo
L
SW1
We may add a small switch to
short the inductor when SW1
and SW2 are both off [Jung 99].
Vdd
C
RL
SW2
Vo
L
SW1
Cx
C
RL
i
Vdd
Vo
Ki
Vx
Vx
16
17. Pseudo-Continuous Conduction Mode
By increasing the size of the ringing suppression switch, a switching
converter may work in pseudo-continuous mode (PCCM). It was first
employed in a single-inductor dual-output (SI-DO) converter to
increase the current handling capability [Ma 03b]. When both SW1
and SW2 are open, the freewheel switch SWFW is closed to allow
free-wheeling of i at Ipccm.
SWFW
L
Vx
SW1
Vdd
Vo
i
SW2
C
i
Ipccm
RL
0
Ki
17
18. Efficiency of Buck Converter
Rs
Idd
Io
L
S1
Vdd
Vo
R
S2
Rd
C
RL
η=
Po
VI
= o o
Pdd VddIdd
For an ideal buck converter working in CCM, the conversion ratio M
is Vo/Vdd = D, and Io:Idd = 1:D, giving η=1. If conduction loss is
accounted for, then Io/Idd is still 1/D, but M is modified as M=ηD,
with
P
1
η= o =
R + DR s + D 'R d
Pdd
1+
RL
Ki
18
19. Efficiency of 2nd Order Converters
By accounting for conduction losses due to switch, diode and inductor
series resistance (Rs, Rd and R , respectively), the efficiencies of buck,
boost and buck-boost converters are computed as [Ki 98]
1
R + DR s + D 'R d
1+
RL
Buck:
Boost:
ηboost =
Buck-boost:
Ki
ηbuck =
ηbuck −boost =
1
1 R + DR s + D 'R d
1+ 2
RL
D'
1
1 R + DR s + D 'R d
1+ 2
RL
D'
19
20. Performance Evaluation Parameters
For a good voltage regulator, the output voltage should remain
constant even the input voltage, load current or temperature changes.
Steady state parameters:
Line regulation
Load regulation
Temperature coefficient
Small signal parameters:
Power supply rejection
Output impedance
Transient parameters:
Line transient (settling times)
Load transient (settling times)
Reference tracking time
Ki
20
21. Line Regulation
Line regulation is the change of Vo w.r.t. the change in Vdd:
line reg. =
=
ΔVo
ΔVdd
in mV / V
ΔVo / Vo
ΔVdd
in % / V
Switching converters are non-linear circuits for large signal changes,
and hand analysis is impossible. It could be obtained by simulation.
In datasheets, line regulation is usually measured.
Ki
21
22. Power Supply Rejection
For a good switching converter (also for bandgap reference and
linear regulator), the output voltage should be a weak function w.r.t.
the supply voltage. Hence, a small signal parameter, the power
supply rejection, gives good indication of line regulation.
Power supply rejection (PSR) is the small signal change of Vo w.r.t.
the small signal change in Vdd.
vo
v dd
In transfer function form:
PSR =
In dB:
PSR = 20 × log
v dd
vo
Usually |vo/vdd| < 1, but we customarily give a positive PSR in dB.
Note:
Ki
Line reg. ≈ PSR × ΔVdd
22
23. Load Regulation and Output Impedance
Load regulation is the change of Vo w.r.t. the change in Io:
load reg. =
=
ΔVo
ΔIo
in mV / mA
ΔVo / Vo
ΔIo
in % / mA
In datasheets, load regulation is usually measured.
In the small signal limit, load regulation is the output impedance:
Ro =
Ki
dVo
dIo
in Ω
23
24. Temperature Coefficient
Temperature coefficient (TC) is the change of a parameter X w.r.t.
the change in T, and is a large signal parameter:
TC =
=
ΔX X(T2 ) − X(T1 )
=
ΔT
T2 − T1
in [X] / o C
ΔX / X
ΔT
in ppm / o C
TC could be positive or negative.
Ki
24
25. PWM Voltage Mode Control (1)
A regulated switching converter consists of the power stage and
the feedback circuit.
MP
Vo
L
Vg
MN
RL
ck
C
R1
CMP
Q
R
Q
va
EA
A(s)
S
va
bVo
Vref
ramp
Q
R2
va
Q
ramp
ck
Ki
For a buck converter, if an on-chip charge pump is not available,
then the NMOS power switch is replaced by a PMOS power switch.
25
26. PWM Voltage Mode Control (2)
The output voltage Vo is scaled down by the resistor string R1 and
R2. The scale factor is b = R2/(R1+R2).
The scaled output voltage bVo is compared to the reference
voltage Vref to generate a lowpass filtered voltage Va through the
compensator A(s).
At the start of the clock, the SR latch is set and the switch MP is
turned on, starting the duty cycle. A sawtooth waveform (ramp)
synchronized with the clock ramps up.
When the ramp reaches the level of Va (trip point), the SR latch is
reset, terminating the duty cycle.
Ki
When the SR latch is set, i ramps up. When the SR latch is reset,
i ramps down. In the steady state, i returns to the same level at
the start of every clock cycle.
26
27. PWM Feedback Action
For stability, the control loop has to have negative feedback.
Assume Vo drops suddenly due to change in load or disturbance
⇒ error voltage Verr = (Vref–bVo) becomes larger
⇒ Va = A(f)(Vref–Vo) also becomes larger
⇒ with a higher Va, it takes the ramp longer to reach Va
⇒ duty ratio D is temporarily increased
⇒ more current is dumped into the load
⇒ Vo rises accordingly and eventually settles to the original
value
Note that A(s) is the frequency response of the compensator,
not of the op amp Aop(s).
Ki
27
28. PWM Current Mode Control
A current mode controlled switching converter is realized by replacing
the fixed voltage ramp with the inductor current ramp.
L
MP
Vo
i
Vdd
RL
MN
C
R1
CMP
Q
R
Q
va
EA
A(s)
S
Vdd
ck
Vref
R2
current
sensor
i /N
va
NR f
Ki
bVo
i Rf
28
29. Sub-harmonic Oscillation and Slope Compensation
Output of EA Va cannot change in one cycle. If inductor current is
perturbed by an amount of ΔI1, oscillation occurs if
ΔI2
−m2
=
> 1 ⇔ D > 0.5
m1
ΔI1
Ia = Va / R f
D < 0.5
Ia = Va / R f
D > 0.5
−m2
m1
ΔI1
ΔI2
ΔI1
m1
−m2
ΔI2
To prevent oscillation, employ slope compensation by adding a negative
slope to Ia (i.e., Va) to suppress the change in ΔI2.
Ia = Va / R f
−mc
m1
Ki
ΔI1
m − m2
m
ΔI2
= c
< 1 ⇔ mc > 2
mc + m1
2
ΔI1
−m2
ΔI2
29
30. Current Mode PWM with Compensation Ramp
In practice, the output of EA (Va) should not be tempered, and a
compensation ramp of +mc is added to m1 instead.
L
MP
Vo
i
Vdd
RL
MN
C
R1
CMP
Q
(m1 + mc )R f
va
Ki
EA
A(s)
S
−(m2 − mc )R f
vb
DT
R
Q
va
bVo
Vref
Vdd
R2
i /N
ck
V2I
vb
NR f
ramp from OSC
compensation
ramp
30
31. Synchronous Rectification
To eliminate loss due to forward diode drop, the power diode is
replaced by a power NMOS MN, and the scheme is known as
synchronous rectification. To eliminate short-circuit loss of MP and
MN, a break-before-make (BBM) buffer is used.
L
MP
Q, VP
i
Vdd
RL
MN
C
VP
VN
BBM
Buffer
Ki
Q
R
Q
S
Additional logic is needed
for DCM operation.
Q
(ck)
VN
φ1
φ2
φ1 =
VP
φ2
Non-overlapping φ1 and φ2
φ1
φ2 = VN
31
32. Multiple-Output Converters
Consider two boost converters that operate in deep DCM:
L
i1
Vdd
Vo1
S1
S0
C1
i1
R L1
T
2T
T
2T
L
i2
Vdd
Ki
Vo2
S2
S0
i2
R L2
C2
32
34. SIMO Converter in PCCM
To handle large load currents, raise the inductor current floor to
operate in PCCM. Add a free-wheeling switch (SFW) to short the
inductor when the inductor current reaches Ipccm [Ma 03b].
SFW
Vo1
S1
L
C1
i
R L1
i
T
S0
Vo2
S2
C2
Ki
2T
T
Vdd
2T
i
R L2
Ipccm
34
35. SI-MIMO Converter
Some applications need two converters in series with reduce efficiency.
Vbat
Vload
Vsrc
Energy-harvesting
source
Boost 1
Rechargeable
battery
Boost 2
Load
Reorganize by using a SI-DIDO converter that needs only one inductor
[Lam 04b], [Lam 07b], [Sze 08].
Vbat
Vbat
Vload
Vsrc
Ki
Energy-harvesting
source
SI-DIDO boost
Load
Rechargeable
battery
35
36. Development of SI-MO and SI-MIMO Converters
The recent years sees active R&D activities of SI-MO and SI-MIMO
switching converters for low power applications. It is important to
recognize the contribution of the first developers.
The idea of SI-MO converters was first conceived in [Goder 97], and
only boost sub-converters were considered.
An SI-DO converter with buck-boost sub-converters was discussed in
[Ma 97] to demonstrate the switching flow graph modeling method.
SI-DO converters became commercial products [MAX 98, UCC 99].
The concept of SI-MO was reinvented [Li 00, Ma 00, Ma 01, May
01]. [Ma 01] stressed the importance of DCM operation for reducing
cross-regulation. A systematic classification is discussed in [Ki 01].
DCM operation is extended to PCCM operation in [Ma 02].
The concept of SI-MIMO was conceived [Lam 04, Lam 07].
Ki
36
37. References: Switching Converter Fundamentals
Books:
[Brown 01] M. Brown, Power Supply Cookbook, EDN, 2001.
[Erickson 01] R. W. Erickson and D. Maksimovic, Fundamentals of Power
Electronics, 2nd Edition, Springer Science, 2001.
[Kassakian 91] J. G. Kassakian, M. F. Schlecht and G. C. Verghese, Principle
of Power Electronics, Addison Wesley, 1991.
[Krein 98] P. E. Krein, Elements of Power Electronics, Oxford, 1998.
Papers:
[Jung 99] S. H. Jung et. al., "An integrated CMOS DC-DC converter for
battery-operated systems," IEEE Power Elec. Specialists Conf.,
pp. 43–47, 1999.
[Ki 98]
Ki
W. H. Ki, "Signal flow graph in loop gain analysis of DC-DC PWM
CCM switching converters," IEEE TCAS-1, pp.644-655, June
1998.
37
38. References: Early Development of SI-MIMO Converters (1)
[Goder 97] D. Goder and H. Santo, “Multiple output regulator with time sequencing,” US
Patent 5,617,015, April 1, 1997.
[Ma 97]
[MAX 98]
"MAX685: Dual-output (positive and negative) DC-DC converter for CCD and
LCD", Maxim Datasheet, 1998.
[UCC 99]
"UCC3941: 1V synchronous boost converter," Datasheet, Unitrode
Semiconductor Products, Jan. 1999.
[Li 00]
T. Li, "Single inductor multiple output boost regulator," US Patent 6,075,295,
June 13, 2000.
[Ma 00]
Ki
Y. H. Ma and K. M. Smedley, "Switching flow-graph nonlinear modeling
method for multistate-switching converters," IEEE Trans. on Power Elec.,
pp.854–861, Sept., 1997.
D. Ma and W. H. Ki, "Single-inductor dual-output integrated boost converter
for portable applications," 4th Hong Kong IEEE Workshop on SMPS, pp. 4251, Nov. 2000.
38
39. References: Early Development of SI-MIMO Converters (2)
[Ma 01a]
D. Ma, W. H. Ki, C. Y. Tsui and P. Mok, "A single-inductor dual-output
integrated DC/DC boost converter for variable voltage scheduling",
IEEE/ACM Asia South Pacific Design Automation Conf., LSI University Design
Contest, pp.19–20, Jan. 2001.
[May 01]
[Ma 01b]
D. Ma, W. H. Ki, P. Mok and C. Y. Tsui, "Single-inductor multiple-output
switching converters with bipolar outputs", IEEE Int'l. Symp. on Circ. and
Syst., pp. III-301 - III-304, Sydney, May 2001.
[Ma 01c]
D. Ma, W. H. Ki, C. Y. Tsui and P. Mok, "A 1.8V single-inductor dual-output
switching converter for power reduction techniques," IEEE Symp. on VLSI
Circ., Kyoto, Japan, pp. 137-140, June 2001.
[Ki 01]
W. H. Ki and D. Ma, "Single-inductor multiple-output switching converters",
IEEE Power Elec. Specialists Conf., Vancouver, Canada, pp.226–231, June
2001.
[Ma 02]
Ki
M. W. May, M. R. May and J. E. Willis, "A synchronous dual-output switching
dc-dc converter using multibit noise-shaped switch control," IEEE Int’l SolidState Circ. Conf., pp.358–359, Jan 2001.
D. Ma, W.H. Ki, and C.Y. Tsui, "A pseudo-CCM / DCM SIMO switching
converter with freewheel switching", IEEE Int'l Solid–State Circ. Conf., San
Francisco, pp.390–391+476. Feb. 2002.
39
40. References: Early Development of SI-MIMO Converters (3)
[Ma 03a]
[Ma 03b]
D. Ma, W. H. Ki and C. Y. Tsui, "A pseudo-CCM/DCM SIMO switching
converter with freewheel switching," IEEE J. of Solid-State Circ., pp. 10071014, June 2003.
[Lam 03]
Y. H. Lam, W. H. Ki, C. Y. Tsui and P. Mok, "Single-inductor dual-input dualoutput switching converter for integrated battery charging and power
regulation," IEEE Int'l. Symp. on Circ. and Syst., Bangkok, Thailand, pp.
III.447-III.450, May 2003.
[Lam 04]
H. Lam, W. H. Ki, C. Y. Tsui and D. Ma, "Integrated 0.9V charge-control
switching converter with self-biased current sensor," IEEE Int'l Midwest
Symp. on Circ. & Sys., pp.II.305–II.308, July 2004.
[Koon 05]
S. C. Koon, Y. H. Lam and W. H. Ki, "Integrated charge-control singleinductor dual-output step-up/step-down converter," IEEE Int'l. Symp. on
Circ. and Syst., Kobe, Japan, pp. 3071-3074, May 2005.
[Lam 07]
Y. H. Lam, W. H. Ki and C. Y. Tsui, "Single-inductor multiple-input multipleoutput switching converter and method of use," US Patent 7,256,568, Aug
14, 2007.
[Ma 09]
Ki
D. Ma, W. H. Ki, C. Y. Tsui and P. Mok, "Single-inductor multiple-output
switching converters with time-multiplexing control in discontinuous
conduction mode," IEEE J. of Solid-State Circ., pp. 89-100, Jan. 2003.
D. Ma, W. H. Ki, and C. Y. Tsui, "Single-inductor multiple-output switching
converters in PCCM with freewheel switching," US Patent 7,432,614, Oct. 7,
2008.
40
41. IC Design of
Power Management Circuits (II)
Wing-Hung Ki
Integrated Power Electronics Laboratory
ECE Dept., HKUST
Clear Water Bay, Hong Kong
www.ee.ust.hk/~eeki
International Symposium on Integrated Circuits
Singapore, Dec. 14, 2009
43. Content
IC Design: Control Loop
Biasing
RTCT oscillator
Comparators, hysteretic comparator
Operational amplifier
Current sensors
Compensation ramp
IC Design: Power Stage
Power transistor and gate drive
Synchronous rectification
Active diodes
Ki
IC Design: Peripheral Circuits
Under voltage lockout (UVLO)
Over current protection (OCP)
Soft start circuit
3
44. Foreword
Analog IC design is ENGINEERING.
Analog IC design is ART.
There is no best design but good and reasonable designs.
Design examples shown are suggestions rather than
instructions, and unavoidably opinionated.
Suggestions and corrections are most welcome to make
a more relevant and accurate presentation.
Ki
4
45. Guidelines for Analog IC Design
(1) Length of transistors are at least 4λ to 8λ for better
matching.
(2) Gate overdrive voltage Vov = (Vgs–Vt) of a transistor
should be at least 150mV for better current mirror
matching.
(3) Use 1% rule as initial point for matching, delay and
losses.
Ki
5
46. PWM Voltage Mode Control (1)
A regulated switching converter consists of the power stage and
the feedback circuit.
MP
Vo
L
Vg
MN
RL
ck
C
R1
CMP
Q
R
Q
va
EA
A(s)
S
va
bVo
Vref
ramp
Q
R2
va
Q
ramp
ck
Ki
For a buck converter, if an on-chip charge pump is not available,
then the NMOS power switch is replaced by a PMOS power switch.
6
47. Current Mode PWM with Compensation Ramp
In practice, the output of EA (Va) should not be tempered, and a
compensation ramp of +mc is added to m1 instead.
L
MP
Vo
i
Vdd
RL
MN
C
R1
CMP
Q
(m1 + mc )R f
va
Ki
EA
A(s)
S
−(m2 − mc )R f
vb
DT
R
Q
va
bVo
Vref
Vdd
R2
i /N
ck
V2I
vb
NR f
ramp from OSC
compensation
ramp
7
48. Synchronous Rectification
To eliminate loss due to forward diode drop, the power diode is
replaced by a power NMOS MN, and the scheme is known as
synchronous rectification. To eliminate short-circuit loss of MP and
MN, a break-before-make (BBM) buffer is used.
L
MP
Q, VP
i
Vdd
RL
MN
C
VP
VN
BBM
Buffer
Ki
Q
R
Q
S
Additional logic is needed
for DCM operation.
Q
(ck)
VN
φ1
φ2
φ1 =
VP
φ2
Non-overlapping φ1 and φ2
φ1
φ2 = VN
8
49. Simple Current Source and Current Mirror
The simple current source is supply dependent. If the power
supply would change considerably, for example, from 5V to 12V,
then the simple current source should not be used.
Vdd
Vdd − V1
1
2
⎛W⎞
I1 = μnC ox ⎜ ⎟ ( V1 − Vtn ) =
2
R1
⎝ L ⎠1
R1
I1
I2
V1
M1
Ki
M2
I2 = I1
≈ I1
(W / L)2 (1 + λn Vds2 )
(W / L)1 (1 + λn Vds1 )
(W / L)2
(1 + λn (Vds2 − Vds1 ))
(W / L)1
9
50. CMOS Widlar Current Source
The self-biased CMOS Widlar current source appears very often in
textbooks. The version with a startup circuit is shown below.
Vdd
long
L 6 M6
M3
V2
M4
I1
I2
V3
M7
M5
wide
W5
startup
Ki
4
R1
:
2
1
⎛W⎞
I2 = μnC ox ⎜ ⎟ ( Vgs2 − Vtn )
2
⎝ L ⎠2
For (W/L)1 = 4(W/L)2 gives
gm1 =
v1i v1o
M1
2
1
⎛W⎞
I1 = μnC ox ⎜ ⎟ ( Vgs1 − I1R1 − Vtn )
2
⎝ L ⎠1
V1
1
M2
2I1
2
=
⇒ gm1R1 = 2
V1 − Vtn R1
A positive loop exists:
−v
−g / g g
−2
TV1 = 1o = m1 m2 m4 =
v1i
1 + gm1R1 gm3
3
10
51. CMOS Peaking Current Source (1)
The original peaking current source was designed in bipolar
processes. [Gray 01] gives a CMOS sub-threshold version, while
we suggest operating all transistors in the active region [Lo 09].
This current source does not need a start-up circuit.
Vdd
M3
Rb
I1
V3
I2
V − Vtn
dI2
= 0 ⇒ I1R1 = 1
dI1
2
R1
V2
Ki
V − V1
1
2
⎛W⎞
μnC ox ⎜ ⎟ ( V1 − Vtn ) = dd
2
Rb
⎝ L ⎠1
1
2
⎛W⎞
I2 = μnC ox ⎜ ⎟ ( V1 − I1R1 − Vtn )
2
⎝ L ⎠2
V1
M1
I1 =
M2
1 : 4
V2
For I2 = I1, set (W/L)2 = 4(W/L)1.
11
52. CMOS Peaking Current Source (2)
The peaking current source has very good power supply rejection.
v dd
Rb
1 / gm3
v1
v3
R1
Ki
R1 =
V1 − Vtn
1
=
⇒ gm1R1 = 1
2I1
gm1
Small signal analysis gives
v2
gm1 v1
From previous analysis,
v 2 1 − gm1R1
=
=0
v dd 1 + gm1R b
gm2 v 2
v dd − v 3 gm2 1 − gm1R1
=
=0
v dd
gm3 1 + gm1R b
That means the currents generated by current mirroring using V2
and (Vdd–V3) has a very low dependence on the supply voltage.
12
53. CMOS Peaking Current Source (3)
Assuming a 0.25µ CMOS process:
Vtn = 0.8V
µnCox = 50µA/V2
|Vtp| = 0.8V
µpCox = 25µA/V2
λnLn = 0.05µm/V
|λp|Lp = 0.05µm/V
Example: Vdd ranges from 4V to 6V, need I2 = 40μA.
Set ∂I2/∂I1 = 0 at Vdd = 5V with I1(5V) = 40μA. (W/L)2 = 40 gives
Vgs2-Vtn = 200mV, and 1mV change in Vtn causes 1% change in I2
(1% rule):
Vtn = 0.799V
Vtn = 0.800V
Vtn = 0.801V
Ki
⇒
⇒
⇒
Vdd
4V
5V
6V
I1
30μA
40μA
50μA
I2 = 40.4μA
I2 = 40.0μA
I2 = 39.6μA
I2
38.7μA
40.0μA
38.9μA
50μA
I1
40μA
I2
30μA
I1
4V
5V
6V
Vdd
13
54. Self-Biased Peaking Current Source
A current mirror (M3, M4) can be used to replace the large Rb to
reduce silicon area. The peaking current source becomes self-biased
and needs a startup circuit (may not be favorable).
Vdd
long
L 6 M6
Vb1
Vb2
M7
M5
wide
W5
Ki
startup
M4
M3
I1
V2
I2
R1
V1
M1
M2
1 :
V1
4
14
55. RTCT Oscillator (1)
The RTCT oscillator generates a ramp that synchronized with the
clock, which fits the requirement of a PWM switching converter.
Vdd
Ich = Vref / R T
Vref
EA
VH
VH
VL
Idch
RT
current
generator
Ki
VC
Mdch
CMP
S
Q
ck
Vm
ramp
T
CMP
CT
VL
T = 1 / fs
R
hysteretic
comparator
15
56. RTCT Oscillator (2)
The charging current Ich is well-controlled by a bandgap derived
voltage Vref and an accurate 1% (external) resistor RT (1% rule).
Ich charges an accurate 1% (external) capacitor CT slowly from the
lower bound VL to the upper bound VH. The ramp excursion is Vm.
The hysteretic comparator trips when VCT > VH, and ck = 1.
When ck = 1, the NMOS Mdch is turned on, and discharges CT with
a large current Idch that is around 10 times of Ich.
When VCT < VL, the comparator trips again, and ck = 0.
Idch is not well-controlled, but the accuracy of the oscillation
(switching) frequency fs is well-controlled because it is dominated
by the accurate Ich.
Ki
16
57. Current Regulator / Voltage Mirror
The error amplifier for generating the charging current can be
realized using a differential amplifier stage.
Vdd
Ich =
Vref
Vref
RT
Ib
Ki
Vref
RT
RT
17
59. 2-Stage Simple Operational Amplifier
The op amp of the PWM compensator should be ground-sensing
(common mode voltage close to ground).
Vdd
M5
Mb
Rb
V−
M1
M7
M2
Ib
R b2
supply
independent bias
Ki
V+
Cc
M3
M4
Vo
Rz
M6
CL
2-stage op amp
19
60. Frequency Response of Op Amp
The op amp gain is Aop(s) (but the EA (compensator) gain is A(s)):
A op (s) =
A dc (1 + s / z1 )
(1 + s / p1 )(1 + s / p2 )
|A|
A dc
p1
where
Ki
A dc = gm1 (rds2 || rds 4 ) × gm6 (rds6 || rds7 )
1
z1 =
C c (R z − 1 / gm6 )
1
p1 =
C c gm1 (rds2 || rds 4 )(rds6 || rds7 )
g
p2 = m6
CL
g
ωt = m1
Cc
choose p2 = 3ωt for φm = 70 o
ωt
p2
z1
/A
ω
ω
−90 o
−180
o
φm
20
61. Current Mirror Amplifier
The simple 2-stage op-amp can be modified to be a 1-stage current
mirror amplifier.
Vdd
Vbp
M0
M7
M8
V−
M5
Rb
4
:1
self-biased
Widlar current source
Ki
A op (s) =
M1
M3
M2
M4
V+
A dc = gm1 (rds6 || rds8 )
Vo
M6
A dc
(1 + s / p1 )
p1 =
1
CL (rds6 || rds8 )
ωt =
gm1
CL
CL
21
62. Folded Cascode Op Amp (1)
To achieve high DC gain, a folded cascode op amp could be used
(assume μn=2μp).
Vdd
8
R b2
:2 :8
Mb6
Mb4
20μA
Vb4
M1
M2
M10
M7
V−
Rb
10μA
M9
M0
Vb3
Mb3
M8
Mb1 Mb5
Vo
V+
Vb2
M5
M6
Vb1
Mb2
Ib
10μA
M3
M4
Ro
CL
Mb0
1 :4
Ki
:4
supply
independent bias
:1
20μA
folded cascode
gain stage
20μA
22
63. Folded Cascode Op Amp (2)
The gain function of the folded cascode op amp is:
A op (s) =
A dc
(1 + s / p1 )
where
A dc = gm1R o
p1 =
1
CL R o
with
R o = [gm6rds6 (rds 4 || rds2 )]|| [gm8rds8rds10 ]
and
ωt =
Ki
gm1
CL
23
64. 2-Stage Folded Cascode Op Amp (1)
To achieve high DC gain, a folded cascode op amp followed by an
inverting stage could be used (assume μn=2μp).
Vdd
8
R b2
:2 :8
Mb6
Mb4
20μA
Rb
M0
M9
M7
V−
Mb2
Ib
Mb1 Mb5
10μA
Vb2
Vb1
M1
M2
Ki
:4
:1
supply
independent bias
M12
M10
M8
Rz C
c
V+
M5
M3
M6
M4
Mb0
1 :4
20μA
Vb4
Vb3
Mb3
10μA
20μA
folded cascode
gain stage
20μA
Vo
R o1
CL
M11
20μA
inverting
gain stage
24
65. 2-Stage Folded Cascode Op Amp (2)
The gain function of the folded cascode op amp is:
A op (s) =
A dc (1 + s / z1 )
(1 + s / p1 )(1 + s / p2 )
where
A dc = gm1R o1 × gm11 (rds11 || rds12 )
R o1 = [gm6rds6 (rds 4 || rds2 )]|| [gm8rds8rds10 ]
1
z1 =
C c (R z − 1 / gm11 )
1
p1 =
C c gm1R o1 (rds11 || rds12 )
g
p2 = m11
CL
g
ωt = m1
Cc
Ki
25
66. V-to-I Conversion for Compensation Ramp
To add a compensation ramp to the inductor current, a V-to-I (V2I)
converter could be used. Two versions are shown below.
Vdd
Vdd
Vin
R
Vin
Ki
V1
Vin
R
Ib
R
Vin
R
(if Vgsn =
| Vgsp |)
V2
Vin
R
26
67. Power Transistor Design
Switch voltage of an ideal switch is 0V when conducting
⇒ MOS switch should have small Vds when conducting
⇒ MOS switch in triode (linear) region when conducting
⇒ For an NMOS power switch MN,
1
⎛W⎞ ⎡
⎤
Id = μnC ox ⎜ ⎟ ⎢(Vdd − Vtn )Vds − Vds 2 ⎥
2
⎝ L ⎠N ⎣
⎦
Vds
1
RN =
=
Id
μnC ox (W / L)N (Vdd − Vtn )
1% rule: conduction loss of RN is 1% of the load RL
If RL is 10Ω, 1% is 100mΩ. If duty ratio is 0.5, then MN conducts
half of the time, and RN can be 200mΩ.
Ki
27
68. Buffer Design
To drive a power switch effectively starting from control logic
blocks, buffers (digital inverters) have to be used.
Minimum delay gives a ratio of e (=2.718), but too many stages
are then needed:
- large transistors give large switching loss;
- large buffers give large shoot-through (short-circuit) loss;
- last stage buffer should have a ratio of 25 to 40.
1
Ki
:
4
:
40
:
600
:
30000
28
69. Eliminate Short-Circuit Loss (1)
For a large inverter, insert a starving resistor Rstarve to limit shootthrough (short-circuit) current, but the most important observation
is at Vp and Vn. For input changes from ‘0’ to ‘1’, Vn drops
immediately, but Vp drops with a delay due to Rstarve.
Vp
Vn
Ki
R starve
Vp
Vn
R starve
29
70. Eliminate Short-Circuit Loss (2)
Short circuit loss of the last stage (largest) buffer could be eliminated
if driven by a buffer with starving resistor.
(W / L)n
=4
40
:
600
40
:
600
Starving resistor can be replaced by transistors operating in the linear
region. A rule of the thumb design is 1/10 of the inverter transistors.
Ki
30
71. Power PMOS or Power NMOS?
The PMOS switch may be replaced by an NMOS switch driven from
an on-chip step-up charge pump [Sze 08].
Vdd = 1.2V
MPS
⇒
VPS
Vref +
−
5X
charge
pump
3.8V
MNS
VNS
level
shifter
Example: Vdd = 1.2V, and needs a 50mΩ switch (RPS = RNS = 50mΩ)
P-switch:
N-switch:
Ki
(W/L)PS = 1/(μpCox×(Vdd-|Vtp|)×RPS)
= 500,000μ/0.25μ
(W/L)NS = 1/(μnCox×(3.8-Vtn)×RNS)
= 33,300μ/0.25μ
Charge pump (1pF caps) + auxiliary circuits is about the size of MNS
⇒
P scheme : N scheme = 7.5 : 1
31
72. Simple P-Current Sensor
On-chip current sensing can be achieved by a small sensing transistor
Mps that is forced to have the same Vd, Vg and Vs as the power
transistor MP using a matched current source [Ki 98].
PVdd 1mA
1
MPs
20 / 2
1A
: 1000
MP
20000 / 2
Q
Q
MSW1
0.999mA
Ms1
to
PWM
CMP
Ki
i R f Ms5
N
Ms2
Ms3
Ms 4
Vo
L
MN
1μA
Rf
AGnd
MSW2
i
Msb
RL
C
PGnd
32
73. Symmetrical Matching of CMOS Transistors
When a pair of transistors M1 and M2 of the same type are matched,
their W/L ratios are the same, i.e., (W/L)1 = (W/L)2, and in most
cases, W1=W2 and L1=L2. However, their drain currents may not be
the same due to channel length modulation.
If in addition to having the same W/L ratio, M1 and M2 are forced
(by an additional circuit) to have essentially the same drain, gate
and source voltages, then they are called symmetrically matched
(SM) [Lam 04b, Lam 07].
The simple current sensor uses the 4T cell to force Mps to be
symmetrically matched with MP. However, the accuracy is limited by
the 4T cell that itself is not symmetrically matched.
Ki
33
74. Symmetrically Matched N-Current Sensor
Replace 4T cell by 8T cell with internal cross-biasing such that paired
transistors (M1, M2), (M3, M4), (M5, M6), (M7, M8) are symmetrically
matched, forcing Vy=Vx. Start-up circuit is needed [Lam04b, Lam 07].
M
Vdd
M9
:
1
M6
Isense
:
1
M4
M3
V4
:
M8
V2
M5
M1
Vx
M
V3
V1
M2
M7
Vy
i
Q
MN
1000
Ki
MNs
:
1
34
75. Concept of Active Diode
The lossy passive diode may be replaced by an active diode, and
eliminate the need to control two switches for synchronous
rectification:
Vx (A)
L
MP
(K)
Vo
(active diode)
RL
MN
Vdd
C
An active diode is simply a power transistor controlled by a (current)
comparator:
A
K
CMP
Ki
A
K
CMP
35
76. Active Diode Implementation (1)
One implementation of the active diode is discussed in [Man 06],
and is used in a DCM boost converter. The capacitor C is added to
improve transient response.
VL (A)
VH (K)
VH
C
Ki
36
77. Active Diode Implementation (2)
Another active diode is used in a regulated charge pump [Lam 06].
VH
VL
VH
VL
VH
VH
Vb
Active diodes can be used as maximum voltage selector for biasing
substrates of PMOS power switches in a multiple-output converter.
V1
Ki
Vmax
V2
37
78. Power Management Peripherals
For a low-voltage system, e.g., Vdd=5V, there is usually only one
trimmed voltage reference.
For a system with Vdd = 15V, there are usually two voltage
references, one trimmed for accuracy, and a second one untrimmed
and could work at very low voltage for start-up, UVLO (under voltage
lockout) and OVP (over voltage protection).
15V
5V
VBG(untrim)
untrim
BGR
UVLO
OVP
+
_
VREF
trimmed
BGR
functional
blocks
linear regulator
Ki
38
80. Soft Start Circuit
Consider the buck converter. When Vo=0, EA drives Va to Vdd, and
D=1. SW1 is always on, causing large in-rush current. The soft start
circuit uses a very tiny current to charge a large CSS, such that VSS
rises very slowly, limiting the duty ratio.
Vdd
ramp
soft
start
Va
VSS
CMP
R Q
ck
R1
M1
M2
S Q
VSS
M2 is in sub-threshold region, sourcing
a current in the range of nA.
Ki
Vbn
CSS
40
81. I/O Connections
Different types of ESD (electrostatic discharge) diodes.
Vdd
I/O
I/O
GND
Schottky
diodes
Ki
large
diodes
diodeconnected
transistors
41
83. Resistor String using Unit Resistors
In an analog circuit array (compared with a digital gate array), all
components are fixed except for the metal layers for
interconnection. Resistors are thus formed using unit resistors.
For example, if the unit resistor is 9.5kΩ and R1=10.8kΩ is needed.
Use continual fraction expansion:
R1
10.8
=
= 1.136842
R unit
9.5
1
=1+
7.3077
1
=1+
7 + (1 / 3.25)
1
=1+
7 + (1 /[3 + (1 / 4)])
Ki
43
84. Parallel/Series Connection of Resistors
For Runit = 9.5kΩ
[1+(1/7)]×Runit = 1.14286×Runit = 10.86kΩ
8 Runit
[1+(1/[7+(1/3)])]×Runit = 1.13636×Runit = 10.8kΩ
11 Runit
[1+(1/[7+(1/[3+(1/4)])])]×Runit = 1.13684×Runit = 10.8kΩ 15 Runit
Use 11 Runit (instead of 15 Runit) is accurate enough and save
components. The structure is:
R1
1
=1+
1
R unit
7+
3
Ki
⇒
R unit
R1
44
85. IC References: Books/Theses
Books / Book Chapters / Thesis:
[Gilbert 96]
B. Gilbert, "Monolithic voltage and current references: Theme and Variations," in
[Huijsing 96], 1996.
[Gray 01]
P. Gray, P. Hurst, S. Lewis and R. Meyer, Analysis and Design of Analog Integrated
Circuits, 4th Ed., Wiley, 2001.
[Huijsing 96] J. H. Huijsing, R. van de Plassche and W. Sansen, Analog Circuit Design, Kulwer,
1996.
[Johns 97]
[Lam 08]
Y. H. Lam, Differential Common-Gate Techniques for High Performance Power
Management Integrated Circuits, PhD Thesis, HKUST, Jan. 14, 2008.
[Meijer 96]
G. Meijer, "Concepts for bandgap references and voltage measurement systems," in
[Huijsing 96], 1996.
[Razavi 01]
B. Razavi, Design of Analog CMOS Integrated Circuits, McGraw Hill, 2001.
[R-Mora 02]
G. A. Rincon-Mora, Voltage References, IEEE Press, 2002.
[Sansen 06]
W. Sansen, Analog Design Essentials, Springer, 2006.
[Sze 81]
Ki
D. Johns and K. Martin, Analog Integrated Circuit Design, Wiley, 1997.
S. M. Sze, Physics of Semiconductor Devices, 2nd Ed., Wiley, 1981.
45
86. IC References: Current Sources
Current Sources and Circuits:
[Frederiksen 72] T. M. Frederiksen, "Constant current source," US Patent 3,659,121, Apr. 25,
1972.
[Lam 07]
[Lo 09]
A. Lo, W. H. Ki and W. H. Mow, “A 20MHz switched-current sample-and-hold
circuit for current mode analog iterative decoders,” IEEE Int’l Symp. on IC, pp.
283-286, 2009.
[Kessel 71]
T. van Kessel and R. van der Plaasche, "Integrated linear basic circuits," Philips
Tech. Rev., pp.1-12, 1971.
[Smith 68]
K. C. Smith and A. Sedra, "The current conveyor: a new circuit building block,"
Proc. of the IEEE., pp.1368-1369, 1968.
[Widlar 65]
Ki
Y. H. Lam, W. H. Ki and C. Y. Tsui, "Symmetrically matched voltage mirrors and
applications therefor," US Patent 7,215,187, May 8, 2007.
R. J. Widlar, "Some circuit design techniques for linear integrated circuits," IEEE
Trans. Circ. Theory, pp.586-590, 1965.
46
87. IC References: Current Sensors and Active Diodes
On-Chip Current Sensors:
[Ki 98]
W. H. Ki, "Current sensing technique using MOS transistors scaling with matched
bipolar current sources," U.S. Patent 5,757,174, May 26, 1998.
[Lam 04a]
H. Lam, W. H. Ki and D. Ma, "Loop gain analysis and development of high-speed
high-accuracy current sensors for switching converters," IEEE Int'l. Symp. on Circ.
& Sys., pp.V.828–V.831, May 2004.
[Lam 04b]
H. Lam, W. H. Ki, C. Y. Tsui and D. Ma, "Integrated 0.9V charge-control switching
converter with self-biased current sensor," IEEE Int'l Midwest Symp. on Circ. & Sys.,
pp.II.305–II.308, July 2004.
[Lam 07]
Y. H. Lam, W. H. Ki and C. Y. Tsui, "Symmetrically matched voltage mirrors and
applications therefor," US Patent 7,215,187, May 8, 2007.
Active Diodes:
[Lam 06]
[Man 06]
Ki
Y. H. Lam, W. H. Ki and C. Y. Tsui, "An integrated 1.8V to 3.3V regulated voltage
doubler using active diodes and dual-loop voltage follower for switch-capacitive
load," VLSI Symp. on Tech. & Circ., pp.104-105, June 2006.
T. Y. Man, P. Mok and M. Chan, "A CMOS-control rectifier for discontinuousconduction mode switching DC-DC converters," IEEE Int'l Solid-State Circ. Conf.,
pp.358-359, Jan. 2006.
47
89. IC Design of
Power Management Circuits (III)
Wing-Hung Ki
Integrated Power Electronics Laboratory
ECE Dept., HKUST
Clear Water Bay, Hong Kong
www.ee.ust.hk/~eeki
International Symposium on Integrated Circuits
Singapore, Dec. 14, 2009
91. Content
Stability and Compensation
Nyquist criteria
System loop gain
Phase margin vs transient response
Type I, II, III compensators
Compensation for voltage mode control
Compensation for current mode control
Ki
3
92. Feedback Systems
Consider the feedback system:
in
F(s)
out
G(s)
Note that F(s) and G(s) are ratios of polynomials in s, that is,
F(s) =
nF (s)
dF (s)
G(s) =
nG (s)
dG (s)
The closed loop transfer function is
H(s) =
out
F(s)
F(s)
=
=
in 1 + F(s)G(s) 1 + T(s)
and the loop gain is
T(s) = F(s)G(s) =
Ki
n(s)
d(s)
4
93. Stability Criteria
Local stability: all poles of T(s) (= all roots of d(s)) are in LHP
System stability:
∗
all poles of H(s) are in LHP
⇒
all zeros of (1+T(s)) are in LHP
⇒
all roots of (n(s)+d(s)) are in LHP
If all functional blocks satisfy local stability, the Nyquist criterion for
system stability is:
∗
Nyquist plot of 1+T(s) does not encircle (0,0)
⇒
Nyquist plot of T(s) does not encircle (-1,0)
If all functional blocks satisfies local stability, the Bode plot criteria
for system stability is:
phase margin φm>0o and gain margin GM>0dB
Ki
5
94. 1st Order Loop Gain Function
T(s) =
Bode Plots
To
|T|
s
1+
p1
To
Nyquist Plot
p1
Im
ωUGF
ω
stable
unit circle
−∞
0
−1
ω = +∞
To
1
-45
o
0−
Re
ω = 0+
To / 2
/A
ω
−45o
−90 o
ωUGF
p1
Ki
6
95. 2nd Order Loop Gain Function
T(s) =
Bode Plots
To
⎛
s ⎞⎛
s ⎞
⎜1 + p ⎟ ⎜1 + p ⎟
⎝
1 ⎠⎝
2 ⎠
|T|
To
p1
p2
Nyquist Plot
ωUGF
Im
ω
stable
/A
−1
φm
ωUGF
0
To
Re
ω
−90 o
−180
Ki
o
φm
7
96. 3rd Order Loop Gain Function
T(s) =
Bode Plots
To
|T|
⎛
s ⎞⎛
s ⎞⎛
s ⎞
⎜1 + p ⎟ ⎜1 + p ⎟ ⎜1 + p ⎟
⎝
1 ⎠⎝
2 ⎠⎝
3 ⎠
To
p1
p2
p3
Nyquist Plot
ωUGF
Im
unstable
ωUGF
(encirclement
of -1)
φm
−1
ω
0
/A
To
Re
ω
−90 o
−180o
Ki
−270o
φm < 0
8
97. Observations on Loop Gain Function
•
1st order systems are unconditional stable.
•
2nd order systems are stable, but a high damping factor would
cause large overshoot and excessive ringing before settling to
the steady state.
•
For 3rd order systems, if the 3rd pole p3 is less than 10X of the
unity gain frequency ωUGF, the system is unstable.
Hence, for a stable system, the loop gain function could be
approximated by a 2nd order loop gain function with the 2nd pole p2
usually larger than ωUGF to achieve small overshoot.
Ki
9
98. Loop Gain Function and Transient Response
The transient response of a feedback system is given by
F(s) ⎞
⎛1
⎛1
⎞
v o (s) = L−1 ⎜ × H(s) ⎟ = L−1 ⎜ ×
⎟
⎝s
⎠
⎝ s 1 + T(s) ⎠
where L-1(⋅) is the inverse Laplace transform of (⋅).
The exact transient response is affected by F(s), however, if only
T(s) is considered, we may consider the modified feedback system:
in
F(s)
out
in'
T(s)
1
G(s)
H(s) =
Ki
F(s)
1 + T(s)
out '
H'(s) =
T(s)
1 + T(s)
10
99. Compensator Considerations
For a loop gain function approximated by a 2-pole function:
T(s) =
To
1
≈
⎛
s ⎞⎛
s ⎞
s ⎛
s ⎞
1 + ⎟ ⎜1 + ⎟
1+ ⎟
⎜
p1 ⎠ ⎝
p2 ⎠ ωUGF ⎜
p2 ⎠
⎝
⎝
The closed loop function (with unity gain
feedback) is
T(s)
1
H'(s) =
=
1 + T(s)
s
s2
1+
+
ωUGF ωUGFp2
Ki
Write H'(s) in standard 2nd order form:
1
ωo = ωUGFp2
H'(s) =
2
1 s
s
1+
+ 2
ωUGF
Q ωo ωo
Q=
p2
|T|
p1
ωUGF
p2
/T
−90 o
−180
o
φm
11
100. Relationship between p2/ωUGF and φm
k
p2
ωUGF
Q
ωUGF
1
=
p2
k
overshoot
e
−
π
2
4Q −1
phase margin
φm = tan−1
1.316
0.275
30 o
1
1
0.163
45o
3 = 1.73
0.76
0.064
60 o
0.605 ≈ 0.6
0.01
70 o
1 / 3 = 0.577
2.73 ≈ 3
Ki
p2
ωUGF
Note that φm=60o gives an overshoot of 6.4%, and the 1% settling
time (tset) would be very long. By setting p2=3ωUGF, then φm=70o,
and the overshoot is only 1%.
12
101. Type I Compensator (0Z1P)
The simplest Type I compensator is an integrator with ωUGF = 1/C1R1.
C1
R1
|A|
Vin
Vout
Vref
Vout
1
= − A(s) = −
Vin
sC1R1
Assume Aop(s) is first order with
ωt >> 1/C1R1:
A op
1
≈
A op (s) =
1 + s / ω1 s / ωt
1
⇒ A(s) ≈
sC1R1 (1 + s / ωt )
Ki
ωUGF =
1
C1R1
ω
ωt
/A
ω
−90 o
−180o
13
102. Type I Compensator (0Z1P)
Type I compensator can be implemented using transconductance
amplifier (OTA). OTA has a very high output resistance ro and
cannot drive resistive loads.
|A|
Vin
Vref
gm
Vout
ro
Using OTA may save one IC pin.
Ki
ωUGF =
Co
Vout
gmro
= − A(s) = −
Vin
1 + sC oro
gmro 1 / C oro
gm
Co
ω
/A
ω
−90 o
14
103. Type II Compensator (1Z2P)
Type II compensator consists of a pole-zero pair with ωz<ωp, and a
maximum phase boosting of 90o is possible.
R2
|A|
C2
1
C2R 2
C1
ωUGF =
1 / C1R1
R1
Vin
Vref
Vout
Vout
1 + sC2R 2
=−
Vin
s(C1 + C 2 )R1 [1 + s(C1 || C 2 )R 2 ]
A(s) ≈
Ki
(1 + sC2R 2 )
sC2R1 (1 + sC1R 2 )
1
C1R 2
(C1 << C2 )
1 / C2R1
ω
/A
ω
90 o phase
boosting
−90 o
15
104. Type II Compensator (1Z1P)
Type II compensator can also be implemented using OTA.
|A|
Vin
Vref
gm
1
C2R 2
Vout
ro
R2
C2
Vout
g r (1 + sC2R 2 )
= − A(s) = − m o
Vin
1 + sC2 (ro + R 2 )
A(s) ≈
Ki
gmro (1 + sC2R 2 )
(1 + sC2ro )
gmro
1
C2ro
ωUGF
g
= m
C2
ω
/A
ω
−90 o
16
105. Type III Compensator (2Z3P)
Type III compensator consists of two pole-zero pairs, and phase
boosting of 180o is possible to compensate for complex poles.
R2
R3
Vin
C3
R1
Vref
|A|
C2
1
C2R 2
C1
Vout
1
C2R1
/A
Vout
(1 + sC 2R 2 )[1 + sC3 (R1 + R 3 )]
=−
+90 o
Vin
s(C1 + C2 )R1 (1 + sC1 || C2R 2 )(1 + sC3R 3 )
(1 + sC2R 2 )(1 + sC 3R1 )
A(s) ≈
sC2R1 (1 + sC1R 2 )(1 + sC 3R 3 )
Ki
(C1 <<C2 , R1>>R 3 )
1
C3R1
1
1
C1R 2 C3R 3
ωUGF
1
=
C1R 3
180o
boosting
ω
ω
−90 o
17
106. PWM Voltage Mode Control
A regulated switching converter consists of the power stage and
the feedback circuit.
MP
Vo
L
Vg
MN
RL
ck
C
R1
CMP
Q
R
Q
va
EA
A(s)
S
va
bVo
Vref
ramp
Q
R2
va
Q
ramp
ck
Ki
For a buck converter, if an on-chip charge pump is not available,
then the NMOS power switch is replaced by a PMOS power switch.
18
107. Loop Gains of Voltage Mode CCM Converters
The system loop gain is T(s) = A(s)×H(s), where A(s) is the frequency
response of the EA (compensator). Loop gains of voltage mode PWM
CCM converters with trailing-edge modulation are compiled. Parasitic
resistances except ESR are excluded [Ki 98].
Buck:
Boost:
T(s) = A(s) ×
bVo 1 + sCR esr
.
sL
DVm
+ s 2LC
1+
RL
bVo [1 − sL / (D '2 R L )]
T(s) = A(s) ×
.
D ' Vm
sL
s 2LC
+
1+ 2
D ' R L D '2
b | Vo | [1 − sDL / (D '2 R L )]
.
Buck-boost: T(s) = A(s) ×
DD ' Vm
sL
s 2LC
+
1+ 2
D ' R L D '2
Ki
19
108. Voltage Mode Compensation (1)
Example: Consider a buck converter with the following parameters:
Vdd=4.2V, Vo=1.8V (D=0.429), Vm=0.5V, b=0.667
L=2μH, C=3.3μF, RL=1.8Ω (Io=1A), Resr=100mΩ, fs=1MHz
The system loop gain is given by
T(s) = A(s) ⋅
5.6 × [1 + s /(3M)]
A(s) × 5.6 × [1 + s /(3M)]
=
1
s
s2
1 s
s2
+
+ 2
1+
1+
2
2.3 390k (390k)
Q ωo ωo
The system loop gain consists of a pair of complex poles, and one
strategy is to use dominant pole compensation.
For a buck converter, the complex pole frequency ωo/2π is 10 to 30
times lower than the switching frequency fs.
Ki
20
109. Voltage Mode Compensation (2)
80
60
Dominant pole compensation
1330 = 62.5dB
A(s) =
1330
1 + s /10
40
20
0dB
390k
H(s) =
5.6 ⇒ 15dB
0.1
1
10
100
1k
10k
100k 1M
5.6 × (1 + s / 3M)
1
s
s2
1+
+
2.3 390k (390k)2
ω
10M
3M
0o
−90
o
ω
/ A(s)
/ H(s)
−180o
Ki
−270o
21
110. Voltage Mode Compensation (3)
80
7500 = 77.5dB
T(s) =
60
7500 × (1 + s / 3M)
⎛
1
s
s2 ⎞
(1 + s /10) ⎜1 +
+
⎟
2.3 390k 390k 2 ⎠
⎝
40
ωUGF
= 75k
20
0dB
0o
−90 o
−180
Ki
o
−270o
390k
ω
0.1
1
10
100
1k
10k 100k
1M
10M
ω
/ T(s)
3M
φm =
90 o
/ H(s)
22
111. Stability inferred from Line and Load Transients
Measuring loop gain could be difficult, and for some circuits, and
especially integrated circuits, due to loading effect and that loopbreaking points may not be accessible, stability is inferred by
simulating or measuring the line transient and/or load transient.
If the circuit is stable and has adequate phase margin, line and load
transients will show first order responses.
If the circuit is stable but has a phase margin less than 70o, line and
load transients will show minor ringing.
If the circuit is unstable, line and load transient will show serious
ringing/oscillation.
Ki
23
112. Current Mode PWM with Compensation Ramp
In practice, the output of EA (Va) should not be tempered, and a
compensation ramp of +mc is added to m1 instead.
L
MP
Vo
i
Vdd
RL
MN
C
R1
CMP
Q
(m1 + mc )R f
va
Ki
EA
A(s)
S
−(m2 − mc )R f
vb
DT
R
Q
va
bVo
Vref
Vdd
R2
i /N
ck
V2I
vb
NR f
ramp from OSC
compensation
ramp
24
113. Loop Gain of Current Mode Buck Converter (1)
The loop gain of a current-mode CCM buck converter with trailingedge modulation is shown below. Others can be found in [Ki 98].
Buck:
1
1
(1 + sCR esr )
CR f n1D ' T
T(s) = A(s) ×
⎛ 1
1 ⎞
1 1 ⎛ 1 (n1D '− D)T ⎞
2
+
+
+
s + s⎜
⎟
CR L n1D ' T ⎟ n1D ' T C ⎜ R L
L
⎝
⎠
⎝
⎠
b×
mc
m
2 −D
, mc > 2 ⇒ n1 >
m1
2
2D '
2L
and R L <
D' T
with n1 = 1 +
The two poles are in general real.
Ki
25
114. Loop Gain of Current Mode Buck Converter (2)
If the poles are real and far apart, the denominator could be
simplified.
Buck:
R L || R a
1 + s / ωz
.
Rf
⎛
s ⎞⎛
s ⎞
1+
1+
⎜
ωa ⎟ ⎜
ωt1 ⎟
⎝
⎠⎝
⎠
m
L
Ra =
n1 = 1 + c
(n1D '− D)T
m1
T(s) = A(s) × b.
ωz =
1
CR c
ωa =
1
C(R L || R a )
ωt1 =
1
n1D ' T
For two real poles that are farther apart, pole-zero compensation
could be used to extend the bandwidth.
Ki
26
115. Current Mode Compensation (1)
Example: Consider a current mode buck converter with the same
parameters as those of the voltage mode converter for comparison.
Vdd = 4.2V, Vo = 1.8V (D = 0.429), b = 0.667, fs = 1MHz, Rf = 1Ω
L = 2μH, C = 3.3μF, RL = 1.8Ω (Io=1A), Resr = 100mΩ, mc = m2
⇒ n1=1.75, 1/n1D’T ≈ 1/1μ
The system loop gain is given by
T(s) = A(s) ×
Ki
0.8(1 + s / 3M)
s ⎞⎛
s ⎞
⎛
1+
1+
⎜
⎟⎜
⎟
290k ⎠ ⎝
880k ⎠
⎝
27
116. Current Mode Compensation (2)
We may assume the poles are far apart and use the simplified
equation, and we have
n1=1.75, Ra=3.5Ω, ωz=3M rad/s, ωa=250k rad/s, ωt1=1M rad/s
The system loop gain is then given by
T(s) = A(s) ×
0.8(1 + s / 3M)
s ⎞⎛
s ⎞
⎛
1+
1+
⎜
⎟⎜
⎟
250k ⎠ ⎝
1M ⎠
⎝
Instead of a pair of complex poles as in voltage mode control, two
separate poles are obtained, and both dominant-pole compensation
and pole-zero compensation could be employed.
Ki
28
117. Current Mode Compensation (3)
80
10000
A(s) =
(1 + s /10)
60
Dominant pole compensation
40
20
250k
0dB
0.8 ⇒ −2dB
H(s)
ω
1M
−20
0o
−90
Ki
o
−180o
3M
ω
1
10
100
1k
10k
100k
1M
10M
/ H(s)
29
118. Current Mode Compensation (4)
80
8000 ⇒ 78dB
T(s) =
60
8000 × (1 + s / 3M)
(1 + s /10)(1 + s / 250k)(1 + s /1M)
40
20
ωUGF
= 80k
0dB
1
10
100
1k
250k
1M
10k 100k
ω
10M
−20
1M
0
o
−90 o
−180
Ki
o
φm
= 70 o
ω
/ T(s)
30
119. Current Mode Compensation (5)
Pole-zero cancellation
60
A(s) =
40
(1 + s / 250k)
(s / 375k)(1 + s / 3M)
375k
250k 3M
20
0dB
ω
1
−20
10
100
1k
10k
H(s)
100k
1M
ω
0o
−90
o
10M
/ A(s)
/ H(s)
−180o
Ki
31
120. Current Mode Compensation (6)
Bandwidth increased by 4 times
to 300k rad/s
60
T(s) =
40
1
(s / 300k)(1 + s /1M)
20
ωUGF = 300k
0dB
1
10
100
1k
10k
100k
−90
−180o
Ki
10M
ω
0o
o
1M
ω
/ T(s)
φm
= 70 o
32
121. References: Switching Converter Compensation
[Brown 01] M. Brown, Power Supply Cookbook, EDN, 2001.
[Ki 98]
[Ma 03a]
D. Ma, W. H. Ki, C. Y. Tsui and P. Mok, "Single-inductor multiple-output
switching converters with time-multiplexing control in discontinuous
conduction mode", IEEE J. of Solid-State Circ., pp.89-100, Jan. 2003.
[Ma 03b]
Ki
W. H. Ki, "Signal flow graph in loop gain analysis of DC-DC PWM CCM
switching converters," IEEE Trans. on Circ. and Syst. 1, pp.644-655, June
1998.
D. Ma, W. H. Ki and C. Y. Tsui, "A pseudo-CCM / DCM SIMO switching
converter with freewheel switching," IEEE J. of Solid-State Circ., pp.10071014, June 2003.
33
123. Voltage Mode Converters: Loop Gain Function
In discussing fast-transient converters, one important parameter
is the loop bandwidth.
The loop gain function of the buck converter with voltage mode
control operating in CCM ignoring ESR is given by [Ki 98]
T(s) = A(s) ×
bVo
.
DVm
1
1+
sL
+ s 2LC
RL
The resonance frequency ωo and the pole-Q are
ωo
=
1
LC
Q
=R
C
L
The converter enters DCM at
R L(BCM) =
Ki
2L
D'T
⇒
QBCM =
2 1
D ' ωo T
35
124. Voltage Mode Converters: Bandwidth Limitation
For voltage mode buck, the ripple voltage is given by
ΔVo
Vo
=
D' 1
8 LCfs 2
If ΔVo/Vo=0.01 and D=0.5, then the complex pole pair is at
ωo
and
= 0.4fs
QBCM
=
2 1
= 10
D ' ωo T
⇒
fo
=
ωo fs
≈
2π 16
To have adequate gain margin GM, say, 6dB, the unity gain
bandwidth fUGF has to be reduced by 10×2=20 times:
fUGF
=
1
f
f
× s = s
20 16 320
If fs=1MHz, then fUGF is at around fs/320 = 3.125kHz.
Ki
36
125. VM Buck: Loop Gain Function with Rδ
The unity gain frequency fUGF of fs/320 is too low. Fortunately (or
unfortunately), the converter inevitably has parasitic resistors
such as RESR, Rℓ (inductor series resistor), Rs (switch resistance)
and Rd (diode resistance), and the loop gain function is [Ki 98]
T(s)
≈ A(s) ×
where
Rδ
bVo
.
DVm
1
⎛ L
⎞ 2
1+ s⎜
+ CR δ ⎟ + s LC
⎝ RL
⎠
≈ R ESR + R + DR s + D'R d
This Rδ is at least 200mΩ, thus reducing QBCM to around 3. With
GM to be 6dB, fUGF is reduced by 3×2=6 times, and
fUGF
Ki
=
1 fs
f
×
≈ s
6 16 100
If fs=1MHz, then fUGF is at around fs/100 = 10kHz.
37
126. VM Buck: Dominant Pole Compensation
|T|
60dB
To
ωp (dominant pole)
-20dB/dec
40dB
20dB
100X
ωUGF ωo
ωs
ω
0dB
GM = 6dB
-60dB/dec
/T
0o
−90 o
−180 o
Ki
−270 o
ω
o
−45 / dec
φm
-180o×Q/dec
38
127. Current Mode Converters: Loop Gain Function
The loop gain function of the buck converter with current mode
control operating in CCM ignoring ESR is given by [Ki 98]
1
1
CR f n1D ' T
T(s) =
⎛ 1
1 ⎞
1 1 ⎛ 1 (n1D '− D)T ⎞
+
+
s2 + s ⎜
⎟ n D' T C ⎜R +
⎟
CR L n1D ' T ⎠
L
⎝
⎝ L
⎠
1
A(s)b ×
with
n1 = 1 +
and
mc
m
2−D
, mc > 2 ⇒ n1 >
m1
2
2D '
R L(BCM) = 2L
D'T
In general, the two poles are real, as discussed next.
Ki
39
128. Current Mode Converters: Bandwidth Limitation
To compute the upper limit of fUGF w.r.t. fs, we simplify the
current mode case as follows. Let D=0.5 and choose n1=2 such
that sub-harmonic oscillation could be suppressed even for
D=0.667. The loop gain function at heavy load is
T(s)
≈ A(s)b
RL
1
R f (1 + sCR L )(1 + sT)
At RL(BCM)=2L/D’T,
TBCM (s) ≈ A(s)b
Ki
R L(BCM)
Rf
1
(1 + s8T)(1 + sT)
Pole-zero cancellation at ω1=1/CRL should be done at the highest
load current Iomax (smallest load resistance). To achieve φm of 70o,
fUGF should be 3 times lower than f2, and fUGF ≈ fs/20. Hence, a
current mode converter could have a unity gain frequency 5 times
higher than its voltage mode counterpart.
40