1. Everything you ever wanted to know about Buck Converter
Design for LED Lighting - unless you are writing a book
Woody Smith
-Design Geek
LINE-IN
PART 1: ANALOG
CONTROLLERS
1
2. 2
Index
Agenda and goals pages 5,6,7
LED drivers first thoughts and specâs pages 8-14
Basic Buck Theory pages 15-19
a) current mode page 20-158
1) First order models pages 21-36
2) Averaged models pages 37-46
3) more accurate models pages 47-158
a)intro pages48-57
b)block diagram CPM transfer functions pages 58-64
c) Tymerski model and transfer funcâs pages 65-67
d) Effect of current feedback on Q pages 68
e) Low Q approx and transfer functions pages 69-77
f)example design pages 78-81
5. AGENDA AND GOALS
1.)Develop a complete design procedure for advanced controller buck converters for
LED Drivers Systems
â˘Focusing on:
â˘Overall power efficiency of the lighting assembly
â˘LED operating efficacy (lumen output per watt of input power)
â˘Thermal management of the LEDs and their driver circuit
â˘AC power factor correction (PFC) for the driver circuitry
â˘AC harmonics generation (distortion)
â˘Meeting EMI restrictions
â˘Whether or not dimming is required
â˘LED driver reliability and service life (to match that of the LEDs)
â˘Circuit protection devices needed
â˘Electronics space efficiency (assembly size)
â˘Cost/competitive position
5
6. AGENDA AND GOALS
Design Approach
1.)Matlab/LTpice macro simulation to handle the majority of the design tasks
--Combined with an Excel spreadsheet for spec and equations
---Showing the strengths of each controller
Current Mode
1.)Peak, Average, OCPM
V2 mode
Sliding Mode
COT
1.)Fixed on time
2.)Fixed off time
Sigma Delta
2.) Comparison of the various architectures
Decision based on the overall system spec
3.) final design in Spectre/Hspice 6
7. AGENDA and GOALS
Design Criteria
â˘LED operating efficacy (lumen output per watt of input power)
â˘Overall power efficiency of the lighting assembly
â˘Thermal management of the LEDs and their driver circuit
⢠only 10-25% of the power is converted to Lumens!!
â˘AC power factor correction (PFC) for the driver circuitry
â˘AC harmonics generation (distortion)
â˘Meeting EMI restrictions
â˘Whether or not dimming is required
â˘LED driver reliability and service life (to match that of the LEDs)
â˘Circuit protection devices needed
â˘Electronics space efficiency (assembly size)
â˘Cost/competitive position
-From an EDN article by
Jim Young, ON Semiconductor, and
Usha Patel, Littelfuse, Inc. - October 8, 2012
7
9. AC dimmer
Optional LED
Microcontroller
AC
Control
(I2C, PWM, Analog)
Voltage,
current,
frequency,
Harmonics, etc.
Monitor
Temperature,
Brightness,
Monitor
GWSnet
Interface
BACnet Bus
Smart Light - Rethink
Customized for building
communication medium
Gateway
Power supply built into the socket
RF
Interface
9
10. AC dimmer
Optional
LED
Microcontroller
AC
Control
(I2C, PWM, Analog)
Voltage,
current,
frequency,
Harmonics,
etc.
Monitor
Temperature,
Brightness,
Monitor
BACnet
Interface
BACnet Bus
Smart Light â Rethink 2
Customized for building
communication medium
Gateway
Power supply built into the socket
RF
Interface
SMPSPFC
Optional
10
11. The following is a primer for a top-level down design
methodology. It emphasizes behavioral modeling and design
oriented analysis(to borrow a phrase from Middlebrook).
Our goal is to shorten the
design cycle and
increase physical insight.
-Chip and board
11
12. First Steps
⢠Whatâs your key specs?
⢠Rank and review them in comparison to existing designs
⢠Build an Excel spreadsheet and project timeline
⢠Review this with your team
⢠Revise and review again
⢠Build a complete system level behavioral model-Matlab and spice
⢠Check all specâs!
⢠Revise your Excel spreadsheet and timeline
⢠Verify your resource needs
⢠List all assumptions
⢠Make sure your Timeline has âguard bands.â
TORA, TORA, TORA!
12
13. Strawman Spec
⢠90% efficiency
⢠50,000 hr. lifetime with greater than 70% of the original luminosity
⢠100W equivalent
⢠Assume 100 lm/W for Leds and 15lm/W for Incandescent -6âish X advantage
⢠Assume Iout=700mA
⢠Vled(700mA) ~ 3.2v (phosphorous coated blue for whiteâish light )
⢠100W/6*700mAá3.2â¤8 leds required
⢠Thermal temp⤠125°C
⢠10-100% dimming Triac based
⢠Triac bleeder ~20mA
⢠Class B EMI spec
13
14. 1. Tmax =125C
2. Tmin= -25C
3. PFC= .9
4. Power eff=.85% min
5. 10-100% dimming
6. Front edge, triac dimmers
7. 20mA bleeder current
8. Vin =100-140V
9. Iout=500mA Âą10ma
10.Vout= 30-32V
11.Inductor ripple=20-40%
12.vout-ripple=?
13.50k hr lifetime
14.Driver dimensions less than
2x3x5mm-working on it
15. driver cost ⤠$1.50 in 100k lots
16. internal OTP, OVP OCP
1. OTP at 125Âą5
2. OVP at input (MOV fuse-
typically at 200V)
3. OCP-TBD
17. class C EN6001
18.UL approved
Specification list
14
15. Ok, letâs get down to it. Letâs start with peak mode Buck theory
âŚâŚâŚletâs start with some solid insights from Abraham Pressman and follow it
with a few(hundred) slides from:
U. of Colorado, ECEN 5807
Great stuff with a solid balance between theory,
design and âŚâŚmath.
15
16. From A. Pressman, Switching Mode
Power Supply , pp 177-178
Fig.
5.5a
Fig. 5.5b
Fig. 5.5c
16
17. From A. Pressman, Switching Mode Power Supply , p 178
KEY
POINTS!!
17
26. 1st Order models work
For small bandwidth systems
Download the Fairchild:
fan4810_http://www.fairchildsemi.com/products/analo
g/fan4810designtools.xls
A wonderful tool for itâs purpose. Use it to explore the
design space.
Read the appnotes and enjoy the wonderful design process, but with your
eyes wide open for itâs limitations
26
27. Read the appnotes; look at pfc setup, check the EMI filtering
There is a ton of things to be learned here-dissect it!!
27
28. This a useful large signal, quasi- linearized model w/o averaging or
small signal considerations.
ic
iL
28
29. Play with the ramps, the compensation. Observe the simplicity of
the switch structure
29
30. 1st order modelâs limitations foster a need for a
more advanced approach.
⢠Start with the large signal models and average
them over the duty cycle phases
⢠Good, but not yet high frequency aware or
sensitive to sampling effects
⢠or linear
⢠Now letâs add State Space matrix formulation and
add small signal perturbations and linearize
⢠Excellent, but still not complete
⢠Time to include sampling-Ridley and gang to the
rescue. But which model?
30
46. Starting with either state space equations or the Tymerski-Vorperian models
you will arrive at this schematic. The only remaining question.: how to model
sampling effects?
46
60. Ignore this for the moment
How cool is this equation. Look how Fm (ie, comp ramp)influences
both loops. Later we will add loop compensation, loop delay, and
sampling effects and it gets really interesting. 60
71. Compare to first-order approximation of the
sampled-data control-to-current model
hfs
s
sT
sT
c
L
sssT
e
esi
si s
s
ďˇď°ďˇďĄ
ďĄďĄ
ďĄ
ďŤ
ď˝
ď
ďŤ
ďŤ
ďť
ď
ď
ď
ď˝
ď
ď
1
1
)/(1
1
1
11
1
1
)(Ë
)(Ë
)/(
1
)/(
1
ď°ďˇ
ď°ďˇ
s
ssT
s
s
e s
ďŤ
ď
ďťď
ď°ď°ďĄ
ďĄ s
a
s
hf
f
m
m
DD
f
f
2
221
1
1
1
ďŤď
ď˝
ďŤ
ď
ď˝
Control-to-inductor current response behaves
approximately as a single-pole transfer function
with a high-frequency pole at
Model (4) is consistent with the sampled-data small-signal model
71
72. A quick aside on
2nd order
transfer function
roots
72
74. Wow, Q< 1 gain is pretty nicely behaved
and the phase variation is getting smooth
74
75. At low Q the poles are separate but interact depending
on their frequency separation(a decade of frequeny to
really separate them phase-wise. Gain-wise decades of
separation removes most interaction)
75
79. Compare to first-order approximation of the
high-frequency sampled-data control-to-current
model
hfs
s
sT
sT
c
L
sssT
e
esi
si s
s
ďˇď°ďˇďĄ
ďĄďĄ
ďĄ
ďŤ
ď˝
ď
ďŤ
ďŤ
ďť
ď
ď
ď
ď˝
ď
ď
1
1
)/(1
1
1
11
1
1
)(Ë
)(Ë
)/(
1
)/(
1
ď°ďˇ
ď°ďˇ
s
ssT
s
s
e s
ďŤ
ď
ďťď
kHz32
221
1
1
1
2
ď˝ď˝
ďŤď
ď˝
ďŤ
ď
ď˝
ď°ď°ď°ďĄ
ďĄ ss
a
s
hf
ff
m
m
DD
f
f
Control-to-inductor current response behaves
approximately as a single-pole transfer function
with a high-frequency pole at 32k vs.39kHz~20%
with Ma=M2. play with
Ma. I grew up using
Ma=( .5-.75) x M2, but
is that opt?
79
81. Example
⢠CPM buck converter:
Vg = 10V, L = 5 mH, C = 75 mF, D = 0.5, V = 5 V,
I = 20 A, R = V/I = 0.25 W, fs = 100 kHz
⢠Inductor current slopes:
m1 = (Vg â V)/L = 1 A/ms
m2 = V/L = 1 A/ms
A/V25.0
2
'
ď˝ď˝
L
TDD
F s
g
D = 0.5: CPM controller is stable for any compensation ramp, ma/m2 > 0
Select: ma/m2 = Ma/M2 = 1, Ma = 1 A/ms
1/A1.0
1
2
1
21
ď˝
ď
ďŤ
ď˝
s
a
m
TMM
M
F
81
86. Repeating the obvious about these models
⢠Averaged models depend on small variations on all variables
⢠Particularly duty cycle
⢠It works well for DCM too, but only if key control variables are well
controlled
⢠Peak current mode wants small âIL to minimize switching uncertainty;
minimum peak to average error(always check this at low current levels)
⢠Low peak/average means low distortion but big inductors
⢠Leading to shallow current ramps and more noise sensitivity at low
vin
⢠Why? Low vin shallow ramps and the same level of
switching noise (comparator switching spikes(~.1-.2v?)
⢠Serious layout worries
86
89. ][Ë)1(]1[Ë][Ë ninini cLL ďĄďĄ ďďŤďď˝
Discrete-time dynamics: )(Ë)(Ë zizi Lc ďŽ
Z-transform: )(Ë)1()(Ë)(Ë 1
zizzizi cLL ďĄďĄ ďďŤď˝ ď
1
1
1
)(Ë
)(Ë
ď
ď
ď
ď˝
zzi
zi
c
L
ďĄ
ďĄDiscrete-time (z-domain) control-to-
inductor current transfer function:
ss TjsT
ee ďˇ
ďĄ
ďĄ
ďĄ
ďĄ
ďď
ď
ď
ďŽ
ď
ď
1
1
1
1
Difference equation:
⢠Pole at z = ďĄ
⢠Stability condition: pole inside the unit circle, |ďĄ| < 1
⢠Frequency response (note that zď1 corresponds to a delay of Ts in
time domain):
89
91. Equivalent hold
⢠The response from the samples iL[n] of the
inductor current to the inductor current
perturbation iL(t) is a pulse of amplitude iL[n] and
length Ts
⢠Hence, in frequency domain, the equivalent hold
has the transfer function previously derived for
the zero-order hold:
s
e ssTď
ď1
ÂŁ[u(t)-u(t)đż(t+Ts)]=
91
92. Complete sampled-data âtransfer
functionâ
s
sT
sT
c
L
sT
e
esi
si s
s
ď
ď
ď
ď
ď
ď˝
1
1
1
)(Ë
)(Ë
ďĄ
ďĄ
2
2
1
2
'
1
m
m
D
D
m
m
mm
mm
a
a
a
a
ďŤ
ď
ďď˝
ďŤ
ď
ďď˝ďĄ
Control-to-inductor current small-signal response:
Ridley, Tan,
Middlebrook-
whoever. The
central issue is
this equation.
How do we
approximate it?
92
93. Example
⢠CPM buck converter:
Vg = 10V, L = 5 mH, C = 75 mF, D = 0.5, V = 5 V,
I = 20 A, R = V/I = 0.25 W, fs = 100 kHz
⢠Inductor current slopes:
m1 = (Vg â V)/L = 1 A/ms
m2 = V/L = 1 A/ms
2
2
2
2
1
2
1
1
'
1
m
m
m
m
m
m
D
D
m
m
mm
mm
a
a
a
a
a
a
ďŤ
ď
ďď˝
ďŤ
ď
ďď˝
ďŤ
ď
ďď˝ďĄ
s
sT
sT
c
L
sT
e
esi
si s
s
ď
ď
ď
ď
ď
ď˝
1
1
1
)(Ë
)(Ë
ďĄ
ďĄ
D = 0.5: CPM controller is stable for any compensation ramp, ma/m2 > 0
93
94. Control-to-inductor current responses
for several compensation ramps (ma/m2 is a
parameter)
10
2
10
3
10
4
10
5
-40
-30
-20
-10
0
10
20
magnitude[db]
iL/ic magnitude and phase responses
10
2
10
3
10
4
10
5
-150
-100
-50
0
frequency [Hz]
phase[deg]
ma/m2=0.1
ma/m2=0.5
ma/m2=1
ma/m2=5
5
1
0.5
0.1
MATLAB file: CPMfr.m
Look how the comp ramp kills
the Ti loop gain and BW
Large Fm approx
94
101. Example
⢠CPM buck converter:
Vg = 10V, L = 5 mH, C = 75
mF, D = 0.5, V = 5 V,
I = 20 A, R = V/I = 0.25 W, fs
= 100 kHz
⢠Inductor current slopes:
m1 = (Vg â V)/L = 1 A/ms
m2 = V/L = 1 A/ms
A/V25.0
2
'
ď˝ď˝
L
TDD
F s
g
D = 0.5: CPM controller is stable for any compensation ramp, ma/m2 > 0
Select: ma/m2 = Ma/M2 = 1, Ma = 1 A/ms
1/A1.0
1
2
1
21
ď˝
ď
ďŤ
ď˝
s
a
m
TMM
M
F
101
112. How Do I Model Current Mode Converters?
I follow a dual Pantheon of âGodsâ
Middlebrook
Erickson
Ridley
Tymerski/Vorperian
Basso
Dixon
Lehman Brooks
Versus
⢠My goal to get a solid physical insight into the operation and modeling of SMPS
⢠Study carefully the approximations in the models
⢠We are taking nonlinear devices and developing :
<averaged> small signal linearized duty cycle controlled models
So you need the math too!
If they have written it, read it!!
112
113. Simon Ang and Oliviaâs book, âPower-Switching converters,â
wonderfully summarizes the process as follows:
This a quite general method that when coupled with the Middlebrook/CĂťk dc-dc
transformer produces insightful physical models and great matlab models
113
114. For Spice ac sim, I head for the Tymerski/Vorperian switching model. Where the
nonlinear switches are replaced with a linearized switching module.
Part of the VPI âmasters of
powerâ group
⢠Assuming <vL>=0
⢠D x Vac =Dâ x Vcp
⢠During D, Va=Vc Vcp=D x Vap
⢠During D,â Vc=Vp Vac=Dâ x Vap
⢠for ton=D, <ia>=D x <ic> and for toff, <ip>=Dâ x <ic>
Applying the small signal variations
⢠IA + ia(ac) =(D + d(ac))(Ic +ic(ac))
⢠Ia(ac)= D x ic
⢠and ip(ac)+ Ip =(Dâ-dâ(ac))(Ic + ic(ac))= Dâx ic(ac) - dâ x Ic
⢠Similarly vcp(ac)= D x vap(ac) + Vap x d(ac)
⢠And vac(ac) =Dâ x vap(ac) âVap x d(ac)
Eq. 1
Eq. 2
Eq. 3
My crude derivation:
114
115. Leading to a simple, elegant model with all the essential physics
It averages, linearizes and embeds AC and DC duty cycle effects into the
switching elements- with bias point sensitivity
115
120. At low Q the poles are separate but interact depending
on their frequency separation(decades of separation to
really separate them phase-wise, but gain-wise only a
decade separation removes most interaction)
120
131. My Quick and Dirty Model for a Peak CPM buck
Recently working at LED lighting house, I inherited an PCPM low-side buck
without an output cap leading to 100% modulation. I needed to rethink my usual
model. Itâs not very original(see the included TI app; and Rendon Hollowayâs
(PET/Nov. 2008, and Ridley) inner-current loop model.
Resistor and gain
Pwm Discrete sampling
+ -
ic^ iL
^
131
132. Woodyâs: the inductor is a current source-sort of current loop model
Nonlinear switching
Ave large
Small signal
perturbation
Linear small
signal
1+
Vin and Freq
sensitive
^
^
Simple 1st order and
not very original, but 132
133. no voltage
loop comp,
min delay Always start with
miminal comp-maybe
add 1/s to get stable,
if needed
Modeling delays
in both loops
133
134. Demo specification
⢠Vin(min)=10v
⢠Vin(max)=15
⢠Vout=5v
⢠Vripple=âv=25mVpp
⢠Vout(droop)=200mV max(Iout changes from 1A to 10A <1us)
⢠Iout(max)=15A
⢠Ts=1us(fs=1Mhz)
⢠Iin-ripple<15mA
On my TODO list
134
135. Low Q(â¤. 5) Approximation:
or
R*C
Quick re-cap of the
formulas for a peak
current buck
135
149. No external
comp ramp
Built this following Brooks Lehmanâs notes
from his course at SantaClara Univ
F1 and vind1 handle the current sensing. Look how clearly it shows the sub-harmonic
instability
149
153. Current Mode decision process
CCM DCM
Pro: Pro:
⢠Lower ripple
⢠Lower peak current
⢠Lower distortion
⢠No sub-harmonics
⢠Faster transient
⢠Less problem light loads
⢠No external ramp
⢠No load dump
Con: Con:
⢠Slower transient
responsible
⢠Light loads eff
⢠Sub-harmonic distortion
⢠Big Caps
⢠Cap stress
⢠High inductor peak current
⢠Bsat increase
⢠Magnetic core losses
⢠Higher wire losses
⢠Mosfet stress increase
153
154. SYNC Non-sync
Pro:
⢠less diode losses
⢠better light load reg
⢠Lower Qrr, Trr losses
Con:
⢠Bigger Switches?
⢠Shoot-thru conduction
⢠Slower response
⢠Without a schottky diode
Pro:
⢠Simpler design
⢠Good at high current
⢠Lower cost(?)
Con:
⢠Higher Thermal
⢠Trr, Qrr losses
⢠Rd losses
Unless you go GaN and
Pay the cost
154
155. Project Design Procedure (part 1)
Questions:
1. What are the Top 3 design criteria
1. Is the design specification approved
2. is the schedule done and approved
1. No, do it now!
2. What are the biggest milestones
3. resources needed
3. Is a custom chip required
1. Process
2. Design tools
3. Chip level spec
4. Any special test or packaging requirements
1. Always check the thermals first and revisit frequently
5. Kick-off meeting w/ the Whole Team
1. As manager, build a complete presentation and discuss it openly
2. if you are a team number be prepared to contribute
3. GET EVERYONE ON BOARD AND COMMITTED
1. Communicate, organize, plan and execute 155
156. Part II: Design- chip/board Top level
1.) Go thru the specification in-depth
a.)clarify and define
2.) build A macro-model
a.) matlab or spice or verilogA- your call
b.) verify itâs correctness
c.) run it thru the spec
1.) where are the hot points?
d.) present the results
3.) Revisit the spec and schedule
a.)revise and get approvals
Part III: Chip Design
Always start with the specification
1.) Is the package(s) defined.
a.) heatsink
b.) wirebonds
2.) latest sim models available
3.) pin out 156
157. 2.) process requirements
a.)latest Cad models
b.) optional layers
3.) layout tools
a.) chip
b.) board
4.) test Requirements
a.) room temp testing only?
b.) trimming
1.) chip
2.) board
c.) Test port needed?
157
158. LED Driver Design Strategy
Letâs assume that we are going to design a CCM, current mode buck,
meeting USA standards with a PWM controller.
Reread the specification doc. For LED lighting with dimming itâs all about
lifetime; efficiency; EMI; PFC(>.9); footprint and full range dimming. Where are
the compromises?
Hint 1: the eye is log sensitive and acutely sensitive to light changes at low
illumination. This says be aware of the need for excellent light load regulation,
clean, smooth startups and full range dimming w/o killing the efficiency.
Hint 2: dimming front edge, rear edge, both?
We will assume front edge(USA typical)
Hint 3: EMI class B
Radiated EMI needs to be checked, but for this task it will be conducted EMI
that is the problem. A buck converter is basically a square wave generator
followed by a filter. Square wave means lots of harmonics distortion; line
filtering leading to possible instability(always include effects of the filters in
your design stability modeling.)
158
159. Hint 4: Lifetime > 50, 000 hours.
Tough, tough, tough. Interplays with cost, performance, size and expensive caps
Hint 5: Is this the base for a multi-product design?.
How are you going to trim and tune it for a range of product specs.
Current limit needs to accurate and fast.
Hint 6: Power supplies for LED driver are about lumens/Watt.
Know your LEDâS! Be aware of lot to lot variations, aging, lumen output versus
temperature
159
160. fc being the current loop unity gain
â
LED driver Design flow
First-cut-no ESR
or ESL
160
161. Letâs look at the equilibrium case now for output cap ripple
given: 1. <IL>=0
2. IL(0)=IL(Ts)
3. Ivalley=Iave-âIL/2 and Ipeak=Iave+âIL/2
4. Iave=Vout/Rload=ILeds
Find the max
Ton=D*Tsw and toff=(1-D)*Tsw
Step 4b contâed
using We arrive at
161
162. Review the fo and Q formulaâs in the ECEN 5807 slides(summary slide 62-63)
Put another way:
Using âvout( last slide) we see
Observe that this formula is kinda âtoplevel-
ishâ with some insight into fo/fsw on ripple
2
2
where
162
165. Rule of Thumb: dv=f(C, fs, Resr, Iripple) For equilibrium voltage ripple approx
Allocate 2/3âs of the allowable ripple to the ESR, and a 1/3 for the cap
Step 5. Calculate the inductor
Inductors are a complete ebook in themselves. I have put a file my dropbox with
an assortment of really interesting reads and it also include class notes from a
couple professors Iâve found enlightening.
Iâm going to start with a less sophisticated approach for my first cut value
165
167. What are the tradeoffs when chosing an Inductor?
ďź More ripple means faster transient response.
--Higher peak current- higher Isat requirements
--More core loss
ďź Typical ripple ratio 20-50% of Iave
ďź Isat should be a minimum 20-30% greater than max Ipeak
ďź packaging
--Cost, parasitic coupling, size
ďź Can the inductor handle load changes fast enough-slew rate
--DCR-resistance
ďź Aging
--Ipeak, Isat,
ďź Switching frequency vs inductor size
--switching losses
My rule
167
168. Also consider for the buck inductor
âIup=âIdown
(Vin-Vout)
2.) V1/L*ton = V2/L*toff
3.) âIave=(âIup +âIdown)/2
1.)
Using 2 and 3 it follows
Lenzâs law
Insights
1.) as L increases, Iripple decreases and Pcore losses decreases and
Isat requirements decrease,
2.) but system response slows too
Voltage cross L:
At Equilibrium:
(Vout)
168
196. -
vref
-
Gci FmGcv
Ëd
Gvd
Gid
Gvg
Gig
Ëvg
Rs
H
vË
ËiL
Average current CPM
The d gain equation is
the key to all these
models for CM
Gcv and Gci tie the voltage and current loops together. The
questions are: how much do they affect each other and which one do
we tune first.
Power
supply
196
211. A first approach
Why do we need an EMI filter?
⢠To stop interference with the power source
⢠HOW? By minimizing the mismatch between the power source
and converter(making resistive)
What are the two conduction modes of EMI?
⢠Common mode
⢠CM EMI is noise relative to the ground plane
⢠Ground return bus or chassis to outputs or supplies
⢠Parasitic cap between driver FET drain and
chassis is a stereotypical killer
⢠Differential mode
⢠Between two outputs or two supplies
⢠Leading cause is the input EMI filter cap impedance
and
⢠Switching devices reverse recovery characteristics
211
212. Undamped filter Buck converter
At low freq |Zemi| is small; but at filter resonance,
|Zemi| can be quite large
simple Input filter
Voltage decreases
Current increases
212
214. Zin(f)
Zemi(f)
No overlap
For Stability
⢠Zin>>Zemi,max
⢠all poles in left plane
⢠EMI filter resonance
frequency in the flat in
the flat band of the
buck input impedance
214
215. Assumptions:
1. At low frequency, a buck converter is constant load.
2. or a current/voltage regulator
3. Or as a neg load resistor from an AC prospective
4. State equation methodology works
P
X1
+
-
X2
Approach 2 -old school
215
216. Old School II- with apologies to my professors
Following the FHA method(fundamental Harmonic analysis) we assume that
the first harmonic has 90% of the energy and it is a sine wave.
⢠Ipeak =2*Iavg(converter 50% duty cycle)
⢠IFH=.636*Ipeak
we find:
216
217. What about resonance peaks?
⢠Time for the damping
⢠And the bench
⢠But first
217
235. Really useful stuff here. Observe the criteria for stability and
for a minimally affected current loop- tied to the load. The
only missing is the frequency and transient responses. We
also need a tool to quickly iterate between filter
modifications
235
236. Middlebrook Extra Element Theorem -EET
Ericksonâs book IS on your shelf-read this section now and then come
back to my excerpts from his class notes at the U. of Colorado
236
237. No problem with
one loop, but what
about 4?
1
Quick re-derive
Added
component
Simple ckt. To
illustrate the
process
Classic
divider
237
239. Now we adjust V1 until vout=0
Nulled!
we adjust itest to null
v(out)â0
Same
as 1
itest
null
Back 2 pages
239
240. The Elegant WayThe Elegant Way
⢠Matlab allows us to do cycle-to-cycle simulation and small signal design
with the same circuit
⢠Add a couple scripts and we can implement Middlebrookâs EET method and
optimize the design with minimal design cycles
⢠Transients, FFT analysis all in the same package
⢠It will be our focal tool for exploring advanced controllers
⢠The following matlab based slides very generated directly from materials
from ECE 5807, U. of Col taught by Prof Dragan Maksimovic. Masterful
Stuff. A serious advantage of running a design group is you get to work with
a group of people from various colleges with different perceptives. Through
them I got introduced to COPE institute writings and dissertations. My next
step is trying to figure out when I can take their Photovoltaic class-hopefully
this summer 240
241. Using state space
equations to model the
converter (any type) and
the filters
SyncBuck-Converter model
PWM âswitching
and small signal
241
243. Driving point input impedance(ZD) and the nulled input impedance(ZN) using the
EET scripts. It is a linearized, averaged, small signal model of the converter
Change the
Rload, and D
and observe
Of course, you could do it in Ltspice âŚâŚ..
243
244. Zoâfilter
output Z Look at their relationships-
peaks separated, but to my
taste, too close in magnitude.
I prefer the filter peak to be
at a lower freq and the ZN to
peak spacing greater than
20db
Investigate the tradeoffs in L versus Cap sizes with parasitics.-Consider space, cost
and aging effects as part of the process
244
245. Looks pretty solid as is, but the
bench and EMI tester will be
the final arbiters
You might want to bode plot just the extra element
addition to the duty cycle to vout transfer function
245
246. Why all the worry about input filters?
[ can ruin a great design]
And make the product unsalable
246
259. Output impedance and caps
Start with the obvious: we assume that the output cap is large
enough to keep output voltage constant- its almost true
IC=iL-IR
The following derivation is
from Daniel Hartâs book
An introductory text with enough
knowledge to do some pretty
smart designs
I derived this
earlier, but for
continuityâŚ.
259
261. C. Basso Circuit with slope
comp and PI voltage comp
Output Impedance with PCM and PI voltage comp
⢠At low frequency Zout(PCM) > Zout(voltage only)
⢠However at LC resonance the game changes
⢠For excessive current loop gain it degenerates back to voltage mode
261
262. Rule of thumb: comp
slope should be 50-75%
Of the off slope in a buck
converter(it is a good
starting pt )
Excessive comp
Looks like std
voltage comp
Why?
262
263. Vin
Ro
AoL
FB1
FB2
⢠R0 is the open loop output resistance of Gain amp with the current loop
⢠FB1 is the current loop
⢠And FB2 is the voltage loop
It is a quick and dirty model, but you can see that feedback lowers
sensitivity, and see how excessive current LG kills the total loop gain
Zoutcl= Zoutopen loop/[1 + Aol* T(loop gain)]
Ideal
gain
Consider a Classic FB setup with output impedance
263
273. Back of the envelope switching loss
Tr
Tr )
Skips more than a little, but it
is a good starting place
1. Add the Tf losses(turn off)
2. Add Rdson losses
And you have a first cut total
loss estimate
Now cut to the simulation
~Vin
{Vin}
273
316. Analogy Between Thermal and Electrical
Resistance
316
Electrical => Q is Charge Thermal => Q is Heat 316
317. What is the Difference Between θ Type vs.
Ď Type Parameter?
θ Type
⢠All the heat flows from the junction to location X
(which remains isothermal)
⢠Location X serves as the external heat sink to the package
Ď Type
⢠Only a fraction of the heat flows from the junction to location X
(non-isothermal)
⢠Temperature gradient exists in location X (different points have different
temperature readings)
(non-isothermal)
317
=Gradient 317
318. Thermals (RÓ¨JA, & RÓ¨JC Definitions)
RÓ¨JA
Thermal impedance from silicon junction to ambient
air temperature. The units of measurement are
ËC/Watt.
RÓ¨JC
Thermal impedance from silicon junction
to device case temperature (all six
sides).
nDissipatio
AJ
θJA
P
TT
R
ď
ď˝
nDissipatio
CaseJ
θJC
P
TT
R
ď
ď˝
318
318
319. Thermal Design Terminology
At each interface from the junction to the ambient
air there is an associated thermal resistance
319
319
320. Thermal Design 101
Typical Values
⢠θCU â 71.4°C/W
⢠Based on 1oz copper, W=1cm
L=1cm. ÎťCU= 4 W/cm K
⢠θVIA â 261°C/W
⢠Based on 0.5oz plating
thickness, for 300um via (12mil)
.
⢠θFR4 â 13.9°C/W
⢠Based on 320um thickness,
W=1cm L=1cm. ÎťFR4=0.0023
W/cm K
⢠θSA â 1000°C/W
⢠W=1cm L=1cm. h= 0.001 W/cm K
320
320
321. Darn Chip Companies Canât Make Up Their
Minds.
θJA is a good metric for thermal performance only
when the board used for measurement is known.
321
Check if it was measured per
Jedec standards.
321
322. Heat Sink Placement
⢠Add the heatsink in
parallel with the
appropriate thermal
resistance based upon
location.
⢠Remember
θSA
â 1000°C/W per cm
squared
θVIA
â (261°C/W) / (# vias)
322
322
323. 323
Design Strategy
1. Determine Max ambient temperature of the application
2. Calculate Internal power dissipation of the IC
â Efficiency, loss elements
3. Make an accurate estimate for required θJA
4. Estimate the Board Heat-sinking Area
TJMAX = (PDMAX * θJA) + TA
323
324. 324
Layout Guideline 1 â Copper Area
⢠Estimate Board Heatsinking size
â Assumes 1oz full copper on top and bottom
â Assumes natural convection
â Assumes ~1W PD
â Assumes infinite thermal vias
BoardArea(in2
) âĽ
77.5 °Cxin2
W
θJA -θJC
BoardArea(cm2
)âĽ
500 °Cxcm2
W
θJA -θJC
BoardArea(cm2
)⼠15.29
BoardArea(in2
) ⼠2.37
xPD
xPD
W
cm2
W
in2
⢠If qJC is known
⢠If unknown
324
325. PCB Area Vs. Junction Temperature
Calculator
PCB
Thermal
Calculator
325
http://www.ti.com/adc/docs/midlevel.tsp?contentId=76735
325
326. Module PCB qJA Examples
⢠AN-2026:
â Curves show qJA vs. Cu area, board size, airflow, heatsink
326
326
328. Solution Size Example
328
328
⢠Max ambient temp = 80C
⢠Max temp rise = 125C â 80C = 45C
Determine allowable Temp rise
Ambient Temp Limited
1
⢠Power requirement = 2.5W (loss, assume
worst case temp)
⢠θJA requirement = 45C / 2.5W = 18C/W
Determine θJA requirement2
⢠PCB size > 3â x 3â
⢠How much board space is available?
⢠Do I need heat sinking or Airflow?
Determine PCB size3
⢠Max allowed PCB = 1.5â square
⢠4-layer θJA = 30C/W
Determine θJA
PCB Size Limited
1
⢠Power requirement = 2.5W (loss)
⢠Temp rise = 2.5W x 30C/W = 75C
⢠Max allowable ambient = 50C
Determine Max Ambient Temp2
⢠What is the applicationâs max ambient?
⢠Airflow or heat sinking may be required
Will it work?3
329. 329
Layout Guideline 2 â Copper Weight
⢠Use thick enough copper
â Example 3âx3â board: 2oz Cu qJA is 25% lower than 1oz
⢠Flood top and bottom with copper if possible
qCu =
Width x Thickness
x Length1
ÎťCu
329
330. 330
Layout Guideline 3 â Thermal Vias
⢠Use lots of vias
â Thermal resistances in parallel
⢠Typical 12mil thermal vias:
⢠Each via:
TC
(TA)
Junction Temperature
TJ __
θVIA
θSAθSAθSAθSA
θSAθSA
θJC
θCu
θCu
θCu
θCu
θCu
θCu
θFR4
θFR4
θFR4
θFR4
θFR4
θFR4
θCu
θCu
θCu
θCu
θCu
θCu
θCu
θVIAS â
# of Thermal Vias
261
°C
W
qVIA =
x Length1
ÎťCu
Ď x [(radius)2
- (radius â plating thickness)2
]
330
331. 331
Layout Guideline 3
⢠More thermal via guidelines in AN-1520
ď qJA vs size, #, and arrangement of vias
â Also qJA vs airflow and PD
This is Must Read! I
originally read as an
Nat Semi Appnote.
Itâs insights will blow
you away.
331
332. 332
Layout Guideline 3
⢠More thermal via guidelines in AN-1520
ď qJA vs size, #, and arrangement of vias
â Also qJA vs airflow and PD
332
368. Letâs talk about output resistance
and Droop under load changes in
voltage and current mode
converters
But first, read this article
368
369. The synopsis: to prevent output droop under load change, we need a
constant, small output impedance under any load and input voltage and a
way to adapt the output voltage
And then read this
369
370. Letâs start with voltage mode
Std Voltage Buck
Small signal voltage Buck
370
371. Small signal Voltage mode Buck during state one
Vâ
Vâ
Zo w/o load
Ăo
+
V
-
IL
Letâs take it to state space now
Buck converter with
load current. For this
analysis I didnât
include the Fet/diode
Rons or back body
Qrr. Easy enough to
do but I a wanted
âcleanâ (i.e., small)
equations
SS input-inductor current
SS input 2
cap voltage
371
372. weThe state space equations for the Buck with output load current as a second
input
Std SS analysis
done in the freq
domain. We will
do it again in the
z domain
372
376. Now letâs look at the current mode controller
tuned for minimum droop
Av
Look at the
Loops
â˘They only share Av
376
377. Great, we can tune the current loop
independent of the voltage loop
â˘Tune Ti first
⢠Note: Ti reduces the overall loop
gain
Whatâs the total loop gain?
G
T2
Reduce the current loop to a Block
â˘We merged Gcv into Gvd
effectively
377
379. What was I expecting you to learn from this
section?
â˘See the usefulness of state space analysis
â˘As an alternative to the Tymerski-Voperian model
â˘See how the loops interact
â˘And now let us simulate a couple examples
379
383. N is the number of phases being
summed at the output cap
Think about this carefully. The Iin-rms (
of a duty cycle dependent square wave)
vs Vo/Rload(DC load current). Now add
in EMI concerns and stability and
parasitics and cost and space
὾=efficiency
D=duty scale
Reasonable starting point, but you really
need to add in the emi filter, layout and
type-sim and bench required
383
447. Synchronous-Buck Controller (1/2)
Synchronous-Buck Controller Circuit with IC
TPS5618 from Texas Instruments
Synchronous-Buck Controller Block Model
(Open Loop Setting)
⢠The Syn-Buck_Ctrl is a block model that generates gate drive pulse signal to control MOSFET
switches of the Synchronous-Buck Converter. The duty cycle, switching frequency, and the
switching dead-time are input into the model to match the real circuit.
HIDR
LODR
High side gate driver
Low side gate driver
447
448. Synchronous-Buck Controller (2/2)
V1
TD = {1/FREQ}
TF = 1n
PW = {D/FREQ}
PER = {1/FREQ}
V1 = 0
TR = 1n
V2 = 1.709
PARAMETERS:
FREQ = 152kHz
D = 0.36
tdly = 80n
0
Rdly 1
1k
N4
CHDR
1nCdly 1
{tdly /1k}
0
0
Rdly 2
1k
N3
0
Cdly 2
{tdly /1k}
HDR
LDR
N1
U1
AND2_ABM
VOH = 12
VOL = 0
Dclmp
DHDR1
U2
AND2_ABM
VOH = 8
VOL = 0
N2
N5
Dclmp
DHDR2
N7
RHDR1
0.01
U5
INV_ABM
VOH = 1.709
VOL = 0
RHDR2
0.01
N6
Pulse
Control
Signal
Dead-time
generator
The Syn-Buck_Ctrl Equivalent Circuit
Parameters
⢠FREQ = Switching frequency, set to match
the measurement switching frequency.
⢠D = Duty Cycle, calculated by DâVOUT/VIN
⢠tdly = HDR and LDR dead-time, the tdly is set
to match the measurement dead time value.
Gate drive signal (measurement)
1/frequency
Dead-time, the time
when QH and QL
are both off
448