Power Distribution Network Design for High-Speed Printed Circuit...
Transcript of Power Distribution Network Design for High-Speed Printed Circuit...
RTPApril 3, 2007
1
Power Distribution Network Design for High-Speed Printed Circuit
Boards
Jun Fan
NCR Corporation
April 3, 2007
RTPApril 3, 2007
2
Outline
• Overview of PDN design in multi-layer PCBs
• Interconnect Inductance
• Individual Capacitor Values
• Capacitor Location
• Power/Ground Plane Pair
• Summary
RTPApril 3, 2007
3
PDN in a Multi-Layer PCB
GND
VCC
DC/DC converter
(Power source)
GNDVCC
IC load
VCCGND
SMT capacitors
VCC
GND
IC driver VCC
GND
electrolyticcapacitor
VCC
GND
GND
IC driver
VDC
switch
Z0, vp
IC load
CL
VCC
RSP-MOS
N-MOS PCB trace
RTPApril 3, 2007
4
Device Switching
IC driver
VCC
RS
shoot-through current GND
Z0, vp
IC load
VCC
charge
logic 0→1
IC driver
Z0, vp
IC load
0 V
discharge
VCC
RS
shoot-through current GND logic 1→0
Current drawn from power
J. L Knighten, B. Archambeault, J. Fan, et. al., “PDN Design Strategies: IV. Sources of PDN Noise,” IEEE EMC Society Newsletter, Winter 2007, Issue No. 212, pp. 66-76.
Shoot-through
current only
Shoot-through + load charging
current
RTPApril 3, 2007
5
Current Supply and Power Bus Noise
VN
VCC
d
IC driver
GND
EMI
EMI
I/O line or cableVnoiseVCC
IC 2
GND
I am Mr. Charge!
RTPApril 3, 2007
6
PDN Design Objectives
1. Ensure charge supply for logic transitions• Enough capacitance to store
charge • Enough charge readily
available for short transitions2. Minimize noise voltage
distribution on the VCC/GND plane pair• Low power bus impedance
over frequency• Noise decoupling• Noise isolation
IC driver
VCC
RS
shoot-through current GND
Z0, vp
IC load
VCC
charge
logic 0→1
VN
VCCd
IC driver
GND
EMI
EMI
I/O line or cableVnoiseVCC
IC 2GND
RTPApril 3, 2007
7
PDN Design Issues in PCB• Capacitor interconnects; • Individual capacitor values and packaging forms;• Number of capacitors needed; • Capacitor placement;• PCB stack-up;• Power/ground plane pair geometry;• Segmentation and isolation
RTPApril 3, 2007
8
PDN Charging/Discharging Hierarchy
The planes can be regarded as a C only @ low frequencies
VDC
Lbulk
Cbulk
Lvia
CSMT
Cplanes
Connectors and wiring Capacitor leads
Electrolytic capacitors
Interconnect
SMT capacitors
VCC/GND planeUltimate Charge Source 100’s uF 0.01-100’s uF pF’s-100 nFVRM (DC/DC converter)
Pin interconnect
BGA packagedIC pins
Speed of charge delivery
Amount of charge available for delivery
RTPApril 3, 2007
9
PDN Impedance Profile
10-4
10-2
100
-50
-40
-30
-20
-10
0
10
20
F req uency [G Hz]
Z 21 [d
BΩ]
Power plane distributed behavior
Global decoupling
Bulk cap
Local decoupling
RTPApril 3, 2007
10
Outline
Interconnect Inductance- shall be minimized in any situation
RTPApril 3, 2007
11
Effect of Inductance
IC loadCLdecoupling capacitor
GNDVCC
Inductance associated with the capacitor: package parasitics, bonding pads, board traces, and vias
Inductance – impedes the flow of current (charge) to the load that will charge it to achieve a logic 1
CjLjRZ
ωω 1++=
LCf res π2
1=
L
C
t=0
v(0)=VS
i(t)
R
)sin()0(LCt
LCVti S=≥
L
C
L
C
R
RTPApril 3, 2007
12
Minimizing Interconnect Inductance
SMT Decoupling Capacitor ESL
Loop 1 Lvia
Loop 2 Labove
The “Good”
The “Bad”
The “Ugly”Really “Ugly”
The “Better” The “Best”
Via
Capacitor Pads
SMT Capacitor
RTPApril 3, 2007
13
Outline
Individual Capacitor Values- multiple values or maximum value? May not matter
LCf res π2
1=
RTPApril 3, 2007
14
Two Common SMT Decoupling Strategies
● Use an array of capacitor values: – This may be the best known approach and is very popular in the
signal integrity design community (SI-LIST).– Rationale: to maintain a flat impedance profile below a target
impedance over a wide frequency range– Typically a logarithmically spaced (10, 22, 47, 100, 220, 470nF,
etc.) array of 3 values per decade.
● Use the largest capacitor value in the package size– This is less well-known, but a popular approach in the EMI
design community– Rationale: to keep impedance as low as possible, less emphasis
on a target impedance and a flat profile
RTPApril 3, 2007
15
-20 dB/dec
+20 dB/dec
decoupling capacitors
distributed behavior
Different Approaches : ComparisonApproach A : values of decoupling capacitors logarithmically spaced, i.e. 3 values per decade : 10, 22, 47, 100, etc.
Approach B : largest values of decoupling available in two package sizes, i.e., 0603 and 0402
Approach B1 : largest values of decoupling available in one package size, i.e., 0402.
Target Impedance
Both approaches can meet the design specs relative to the target impedance
RTPApril 3, 2007
16
Outline
Capacitor Location- could be important due to PCB geometry
RTPApril 3, 2007
17
Background
• Does capacitor location matter?
• How close is close?• What if capacitors can’t
be placed close enough to the IC pwr/gnd pins?
• Is it worth sacrificing routing or using costly new technology in order to get capacitors closer?
• Need a way to facilitate engineering judgment
GND 1.5V 3.3V
decoupling caps
RTPApril 3, 2007
18
Location Affects Interconnect Inductance
Interconnect inductance is reduced when the cap is moved closer to the IC ⇒ The capacitor becomes more effective
CjLjRZ
ωω 1++=
LCf res π2
1=
L
C
t=0
v(0)=VS
i(t)
R
)sin()0(LCt
LCVti S=≥
CAP IC
Lbelow
i(t)
P
G
Labove Labove
CAPS
S
L
C
L
C
R
RTPApril 3, 2007
19
Equivalent Circuit Model
AB
CD
1L2L
2V1V
k
1IParallel Plate
Geometry
Port 1 Port 2
+ +
decCNI3 3L L ESL′= +
J. Fan, et. al., ``Quantifying SMT decoupling capacitor placement in DC power bus design for multi-layer PCBs,’’ IEEE Transactions on Electromagnetic Compatibility, Vol. 43, No. 4, pp. 588-599, November 2001.
RTPApril 3, 2007
20
AB
CD
1L2L
2V1V
k
1IParallel Plate
Geometry
Port 1 Port 2
+ +
decCNI3 3L L ESL′= +
Local Decoupling Effect – Frequency-Domain
' 221
1
VZI
=
2
3
2
3
1132
321
21
21
1
)1(
)()(
LL
LLk
ZLLjMLLj
II
ZZ
N +
+−≈
′++−+==
′ ωω
Equivalent circuit at high frequencies2
21N
VZI
=
' 111
1
VZI
=AB
CD
1L M−
2V1V
1IParallel Plate
Geometry
Port 1 Port 2
+ +
NI
2 3L L M+ −
M
decC
AB
CD
1L M−
2V1V
1IParallel Plate
Geometry
Port 1 Port 2
+ +
NI
2 3L L M+ −
M
01 →decCω
RTPApril 3, 2007
21
Local Decoupling Effect – Frequency-Domain
cfΔ
ZΔ
ZΔ
ZΔ
• Increasing series resonant frequency of decoupling capacitor
• Reducing impedance uniformly in a frequency-independent manner (approximately) for frequencies higher than the series resonant frequency
VCC
GND
VCC IC
VCC
VCC GND
RTPApril 3, 2007
22
Local Decoupling Effect – Time-Domain
M
L2+L3-M
L1-M
iN(t)
i1(t)
i2(t)dt
tdiMLLdt
tdiM )()()( 232
1 −+≈
Equivalent circuit at early time (t≈0)
)()(32
321 ti
LLMLLti N+
−+=
)(1
)1(
2
3
2
3
ti
LL
LLk
N
+
+−≈1L M−
2V1V
1IParallel Plate
Geometry
Port 1 Port 2
+ +
NI
2 3L L M+ −
M
decC
RTPApril 3, 2007
23
Local Decoupling Effect – Time Domain
Reducing initial noise voltage generated in the power bus due to logic transitions.
Time (ns)
Port
2 vo
ltage
(mV
)
VCC
GND
VCC IC
VCC
VCC GND
RTPApril 3, 2007
24
Local Decoupling Effect – Charge Delivery
0 1 2 3 4 5 62
2.2
2.4
2.6
2.8
3
3.2
3.4
time [ns]
Vpl
ane
[V]
50 mils400 mils5000 milsNo decap
0 1 2 3 4 5 62
2.2
2.4
2.6
2.8
3
3.2
3.4
time [ns]
Vpl
ane
[V]
50 mils400 mils5000 milsNo decap
35 mil thick 10 mil thick
RTPApril 3, 2007
25
Estimating Local Decoupling Effect
02 ln 0.75
2equivRdL Hr
μπ
⎡ ⎤⎛ ⎞= −⎢ ⎥⎜ ⎟
⎝ ⎠⎣ ⎦
ln 0 .75
ln 0 .75
equiv
equiv
Rs r
kR
r
⎛ ⎞−⎜ ⎟+⎝ ⎠≈
⎛ ⎞−⎜ ⎟
⎝ ⎠
4equiva bR +≈For a rectangular power bus,
s – spacing between two vias
d – PWR/GND plane pair spacing r – via radiusRequiv – equivalent radius of power bus
Closed-form expressions derived from the radial transmission-line theory :
assumption: vias are located close to the center of the planes
accounting for fringing effect
⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢
⎣
⎡
+
+−≈Δ
2
3
2
3
1021
1
)1(log20)(||
LL
LLk
dBZ
J. Fan, et. al., “Estimating the noise mitigation effect of local decoupling in printed circuit boards,” IEEE Transactions on Advanced Packaging, Vol. 25, No. 2, pp. 154-165, May 2002.
RTPApril 3, 2007
26
220 ln 1 1
2s s
ps ss s
h h l lM hl l h h
μπ
⎧ ⎫⎡ ⎤ ⎛ ⎞⎪ ⎪⎛ ⎞⎢ ⎥= + + + − +⎨ ⎬⎜ ⎟⎜ ⎟⎢ ⎥⎝ ⎠ ⎝ ⎠⎪ ⎪⎣ ⎦⎩ ⎭
220 ln 1 1
2s s
ps ss s
h h r rL hr r h h
μπ
⎧ ⎫⎡ ⎤ ⎛ ⎞⎪ ⎪⎛ ⎞⎢ ⎥= + + + − +⎨ ⎬⎜ ⎟⎜ ⎟⎢ ⎥⎝ ⎠ ⎝ ⎠⎪ ⎪⎣ ⎦⎩ ⎭
( )2via ps psL L M= −
( )
( )
( )
( )
2 2 2 2 2 2
2 2 22 2
2 2 2
2 2 2 2
2 2 272
2 2 2 2
1
2 2 2
2 2 2 2
2 2 32
2 2
2 23
ln
2 23
10 ln
4 tan
2 23
4ln3
t
w p l w p l
l w p ll w p
l w p l
w p w p
w w p lM l w
w w w p l
wlwplp w p l
l p l p
l l p pp ll l p
−
−
⎡⎢− − + + +⎢⎢
⎛ ⎞⎢ + + +⎜ ⎟+ −⎜ ⎟− + + +⎝ ⎠
+ − +
⎛ ⎞+ + +⎜ ⎟= +⎜ ⎟− + + +⎝ ⎠
⎛ ⎞⎜ ⎟− ⋅⎜ ⎟+ +⎝ ⎠
+ − +
⎛ ⎞+ +⎜ ⎟+ +⎜ ⎟− + +⎝ ⎠⎣
⎤⎥⎥⎥⎥
⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎦
( )
2 2 2 22 2
7
23/ 22 2 3 3
3 ln 3 ln2 10 3t
l w l w w lw l l ww lL
ww l l w
−
⎡ ⎤⎛ ⎞ ⎛ ⎞+ + + +⎢ ⎥+⎜ ⎟ ⎜ ⎟× ⎜ ⎟ ⎜ ⎟⎢ ⎥= ⎝ ⎠ ⎝ ⎠⎢ ⎥⎢ ⎥− + + +⎣ ⎦
trace t tL L M= −3 via traceL L L′ = + where and
l: separation between two via; hs: height of the capacitor from the nearest power
or ground plane ;w: width of the trace connecting two vias, or width
of the capacitor package if no trace ;r: radius of the via ; and p = 2hs.
Estimating Local Decoupling Effect
C. Wang, et. al., “An efficient approach for power delivery network design with closed-form expressions for parasitic interconnect inductances,” IEEE Transactions on Advanced Packaging, Vol. 29, No. 2, pp. 320-334, May 2006.
Validated with
CEMPIE
cap
pad
via
trace
RTPApril 3, 2007
27
ICCAP
GNDPWR
IC
CAP
GNDPWR
Dos Don’ts
Design Implications
J. L Knighten, B. Archambeault, J. Fan, et. al., “PDN Design Strategies: II. Ceramic SMT Decoupling Capacitors – Does Location Matter?,” IEEE EMC Society Newsletter, Issue No. 207, Winter 2006, pp. 56-67.
RTPApril 3, 2007
28
Design ImplicationsThe ratio of L3/L2• A good indicator of the ability of a power bus to support local decoupling • The lower, the better• Can be greatly increased by even very short trace length from via to pad
PWR/GND pair
thickness L3' (nH)
L3/L2with no
trace
L3/L2 w/extra 100 mil trace
length
L3/L2 w/extra 200 mil trace
length
L3/L2 w/extra 300 mil trace
length10 mils 1.66 6.75 9.13 11.50 13.8835 mils 0.92 1.29 1.98 2.67 3.36
Tabulated values for • Centered power bus in a 62-mil PCB
stack-up• 10 mil via diameter• 0603 SMT capacitor package
cap
pad
via
trace
RTPApril 3, 2007
29
Design Implications
SMT de-cap Placement L3' (nH) L3/L2
Obverse 0.45 2.96875
Reverse 2.56 9.5625
The placement of de-cap (obverse or reverse), can be highly dependent on the position of power bus in PCB stack-up.
Always LEAST above-plane inductance!
RTPApril 3, 2007
30
Power/Ground Plane Pair- thin is always better
Outline
RTPApril 3, 2007
31
Pi Pj
Impedance of the Power/Ground Plane Pair
dZij ∝)(ωC. Wang, et. al., “An efficient approach for power delivery network design with closed-form expressions for parasitic interconnect inductances,” IEEE Transactions on Advanced Packaging, Vol. 29, No. 2, pp. 320-334, May 2006.
( ) ( ) ( ) ( )⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡+
−+=
≠== =∑∑ HM
ijyjyixjxi
nm
M
mjiji
N
n nm
mn
planeij LjLLLLpyyxxf
abkkdj
CjZ ωεεωμ
ωω ,,,,,,1
00 0
22
22
Location Port Dimensions
Higher order mode L
Resonant frequency contributions
Low frequency capacitive term
exists only when i = j
Based on the cavity method:
RTPApril 3, 2007
32
Power Bus Thickness Effects on Decoupling
1 10 100 1000 5000-50
-40
-30
-20
-10
0
10
20
30
40
Frequency (MHz)
|Z11
| (dB
Ω)
d = 1 mild = 5 milsd = 10 milsd = 40 mils
1 10 100 1000 5000-80
-60
-40
-20
0
20
40
Frequency (MHz)
|Z21
| (dB
Ω)
d = 1 mild = 5 milsd = 10 milsd = 40 mils
Frequency domain: impedance
a = 12 in
b = 10 in Port 2
Port 1(3 in, 2 in)
(7 in, 4 in)d
RTPApril 3, 2007
33
a = 12 in
b = 10 in Port 2
Port 1(3 in, 2 in)
(7 in, 4 in)d
Power Bus Thickness Effects on Decoupling
0 0.5 1 1.5 2 2.5 3-100
0
100
200
300
400
500
Time (ns)
Port
2 vo
ltage
(mV
)
d = 1 mild = 5 milsd = 10 milsd = 40 mils
0 0.5 1 1.5 2 2.5 3-4000
-3000
-2000
-1000
0
1000
2000
3000
4000
Time (ns)
Port
1 vo
ltage
(mV
)
d = 1 mild = 5 milsd = 10 milsd = 40 mils
Time domain: port voltages
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
Time (ns)
Cur
rent
(A)
port 1 current
RTPApril 3, 2007
34
0 1 2 3 4 5 62
2.2
2.4
2.6
2.8
3
3.2
3.4
time [ns]
Vpl
ane
[V]
L3 0 .5nHL3 1nHL3 2nHL3 3nHNo decap
0 1 2 3 4 5 62
2.2
2.4
2.6
2.8
3
3.2
3.4
time [ns]
Vpl
ane
[V]
L3 0.5nHL3 1nHL3 2nHL3 3nHNo decap
Power Bus Thickness Effects on Decoupling
35 mil thick 10 mil thick
RTPApril 3, 2007
35
Effect of Typical Laminates on Decoupling
Dk Loss Tangent
Thick-ness, d
Comment
4.4 0.02 4, 10, 35 mils
better quality fiberglass/ epoxy resin material
4 0.02 2 mils patented thin FR4 core, widely used.
16 0.006 16 microns (0.63 mils)
ultra-thin material commercially available
20 0.015 0.2 mils ultra-thin material from [7, 8].
10 100 1000 5000-60
-50
-40
-30
-20
-10
0
10
20
30
40
Frequency (MHz)
|Z11
| (dB
Ω)
εr = 4.4, tanδ = 0.02, d = 4 milsεr = 4.4, tanδ = 0.02, d = 10 milsεr = 4.4, tanδ = 0.02, d = 35 milsεr = 4, tanδ = 0.018, d = 2 milsεr = 16, tanδ = 0.006, d = 0.63 milsεr = 20, tanδ = 0.015, d = 0.2 mils
Embedded capacitance (thin laminate) has superior electrical performance
RTPApril 3, 2007
36
Summary
Interconnect inductance shall be minimized in any situation: inductance limits the effectiveness of decoupling
Individual capacitor values may not matter: using an array of capacitor values or the largest value in a package size meetsdesign specifications
Capacitor location could be important due to PCB geometry: thin power bus structures (< 10 mils) usually result in a larger L3/L2 value; thus capacitor location is relatively unimportant. Thick power bus structures usually result in a smaller (L3/L2) value, and capacitor placement can be an important design factor
Thin power/ground plane pair is always better: low impedance and high charge availability