Upload
idmond
View
219
Download
0
Embed Size (px)
DESCRIPTION
PCB Design in the GHz Age
Citation preview
McNEILL: PCB Design in the GHz Age 1
PCB Design in the GHz Age
John McNeillWorcester Polytechnic Institute
McNEILL: PCB Design in the GHz Age 2
Overview
• Introduction: Faster / Hotter / Smaller• Problems: Implications for PCB Design• Now What? Advice for ...
–Academia–Industry–Individuals
• Conclusion
McNEILL: PCB Design in the GHz Age 3
IC Design Trends Affecting PCB Design
IC Design PCB Design
INC
RE
AS
ING
DE
CR
EA
SIN
G
SIGNAL SPEED,
EDGE RATE
PIN COUNT
CONTACTSPACING
SUPPLYVOLTAGE
ELECTRICAL THERMAL MECHANICAL
NOISEMARGIN
MOSFETGATE
L<< 1µm
POWERDISSIPATION
SYSTEM ON A CHIP
COMPLEXITY
PACKAGECOMPLEXITY
WAFERSIZE
⇒ 30cm
McNEILL: PCB Design in the GHz Age 4
Microprocessor Evolution
Sources:Brey, The Intel Microprocessors, Prentice Hall2001 ISSCC Digest of Technical Papers, IEEE
Vendor µP Year Pins Package V I [A] Bus [MHz]Intel 80C86 1981 40 DIP 5.0 0.01 5Intel 186 1981 68 PGA 5.0 0.06 25Intel 386 1986 132 PGA 5.0 0.6 33Intel 486 1989 168 PGA 5.0 1.2 33Intel Pentium 1993 237 PGA 3.3 3.4 66Intel Pentium Pro (P6) 1995 387 PGA 2.7 9.9 66Intel Pentium II 1998 242 module 1.8 14 100Sun MAJC 2001 1879 Flip-chip CCGA 1.8 10 400IBM POWER4 2001 5184 module 1.5 77 500
McNEILL: PCB Design in the GHz Age 5
Implications for PCB Design
• Faster Edge Rates⇒ Transmission Line, Signal Integrity Issues
• Higher Current Requirements⇒ Thermal, dI/dt Issues
• Increased Package Complexity⇒ Mechanical, Manufacturability Issues
• Life Getting More Difficult Challenging Interesting
McNEILL: PCB Design in the GHz Age 6
Power Requirements
• Large dI/dt ⇒ Low inductance supply current paths
• V, I extrapolate to V=0, I=∞ in year 2010 ...
1
10
100
0.0 2.0 4.0 6.0
Series1
1989
1993
2001
19951998
2001
SUPPLY VOLTAGE
SUPPLYCURRENT
McNEILL: PCB Design in the GHz Age 7
Faster Edge Rates
• Transmission Line Effects• Rise time x Propagation velocity < Trace length
tr = 200 ps
tr = 2 ns
5 cm
McNEILL: PCB Design in the GHz Age 8
Reflection Example
• Clock Distribution Network
15 cm 15 cm
V1
Z=50Ω
Z=50Ω
Z=50Ω
Z=50Ω
Z=50Ω
McNEILL: PCB Design in the GHz Age 9
Reflection Example: tr ≈ 10ns
Transmission line effects negligible
1 0 8 0 7 3 0 10 210 30. sec . . secE E cm cm cm
RISE TIME VELOCITY
− [ ]( ) ⋅ ⋅ + [ ]( )( ) = >1 244 344 1 24444 34444
McNEILL: PCB Design in the GHz Age 10
Reflection Example: tr ≈ 200ps
• Transmission linereflection problems:–Multiple clocking–Confusion formultilevelsignaling(e.g. Rambus®)
2 0 10 0 7 3 0 10 4 2 30. sec . . sec .E E cm cm cm
RISE TIME VELOCITY
− [ ]( ) ⋅ ⋅ + [ ]( )( ) = <1 244 344 1 24444 34444
McNEILL: PCB Design in the GHz Age 11
Crosstalk / Jitter
• "Digital" Logic, 0 or 1Implication: only level (high/low) is important, but...
• Edge (0 to 1 transition) also matters: time information!• Effect of crosstalk = "jitter"
CROSSTALK∆V
JITTER∆t
"VICTIM"SIGNAL
"AGGRESSOR"SIGNAL
McNEILL: PCB Design in the GHz Age 12
Crosstalk / Jitter Experimental Configuration
• Models poor layout of long run on backplane• Note that even “low frequency” interfering signal can
cause problems ...
60 cm VR
1GHz
10MHz
McNEILL: PCB Design in the GHz Age 13
Crosstalk / Jitter Example: Baseline
Baseline jitter (adjacent line inactive) = 1.5ps rms
McNEILL: PCB Design in the GHz Age 14
Crosstalk / Jitter Example
Jitter with crosstalk (adjacent line active) = 8.6ps rms
McNEILL: PCB Design in the GHz Age 15
Now What?
• Education (antidote to fear, superstition)
Advice for …• Academia
–Reach out to student needs where they are• Industry
–Create culture to support education• Individuals
–Honest self-reflection
McNEILL: PCB Design in the GHz Age 16
One Problem in Academia: Students
• In GHz age, everyone is an analog designer–Never mind Karnaugh maps–Digital designers need awareness of E&M fields
• Problem:–Students don't know what they need to know–Professors must "sell" unglamorous but necessarymaterial
• WPI motto–Old: "Lehr und Kunst" (Theory and Practice)–New: "Someday you'll thank us"
McNEILL: PCB Design in the GHz Age 17
Other Problem in Academia: Professors
• Most professors are freaks geniuses–The ones who understood everything the first time–"Pure beauty of Poisson's theorem”–The rest of us had no idea what was going on
• Present information, not motivation–Expect students to pay attention for material only
• Result: to get good grades, students memorizemathematical tools/techniques but miss meaning
For Example ...
McNEILL: PCB Design in the GHz Age 18
Example
• Nondestructivetest research
• Inject current intometallic sample
• Measure voltagedistribution onsurface
• My expectedprofile
V
x
I
x
McNEILL: PCB Design in the GHz Age 19
Example
• Student's prediction• Ripples "from the
sine waves"–Tried to solvedifferentialequation(Poisson's)with sinedecomposition
BUT:Sine waves aren’t
really there!
V
x
I
x
McNEILL: PCB Design in the GHz Age 20
Real Meaning of Poisson's Equation
• Poisson's equation• 2-d (surface of sample)• ρ=0 (no net charge)
x-∆x x x+∆x
V(x-∆x)
V(x)
V(x+∆x)
d V
dx
V x x V x
x
V x V x x
xx
d V
dx
V x x V x x V x
x
UPPER SLOPE LOWER SLOPE
2
2
2
2 22
≈
+ − − − −
≈ + + − −
( ) ( ) ( ) ( )
( ) ( ) ( )
" " " "
∆∆
∆∆
∆
∆ ∆∆
6 7444 8444 6 7444 8444• Discrete approximation ofsecond derivative(Change in slope per ∆x)
∇ = ⇒ + =22
2
2
20 0Vd V
dx
d V
dy
McNEILL: PCB Design in the GHz Age 21
Result
• V at any point is average of"nearest neighbor" voltages
• Voltage varies smoothly–Note: leads into finiteelement analysis!
V x y V x y V x y V x y V x y V x y
V x yV x y V x
d V dx d V dy
( , ) ( , ) ( , ) ( , ) ( , ) ( , )
( , )( , ) (
+ + − − + + + − − =
= + + −
∆ ∆∆
∆ ∆∆
∆
2 202 2
2 2 2 26 7444444 8444444 6 7444444 8444444
∆∆ ∆ ∆, ) ( , ) ( , )y V x y V x y+ + + −4
V(x-∆, y) V(x, y) V(x+∆, y)
V(x, y-∆)
V(x, y+∆)
McNEILL: PCB Design in the GHz Age 22
Engineering Education Dilemma
• First: courses in mathematical tools–Prove truth, within a formal deductive system–Illusion of certainty
• Then: courses in engineering–Math is a description language for the world–Approximate description of reality
• Reality: World first, mathematical description second
McNEILL: PCB Design in the GHz Age 23
Science ≠ Truth
Do you swear to tell the truth …
"The thought crossed my mind that in science the "truth"does not exist, since science does not prove but canonly describe the ways of nature.”
– Mario Salvadori, "Why Buildings Fall Down", WW Norton, New York, ISBN 0-393-31152-X
McNEILL: PCB Design in the GHz Age 24
Science ≠ Truth
Do you swear to tell the truth …
"The thought crossed my mind that in science the "truth"does not exist, since science does not prove but canonly describe the ways of nature.”
"I answered 'I do' anyway ...”
– Mario Salvadori, "Why Buildings Fall Down", WW Norton, New York, ISBN 0-393-31152-X
McNEILL: PCB Design in the GHz Age 25
Significance for Engineering
• FIRST: Qualitative insight into behavior“Voltage varies smoothly”
–Need to guide design–Choose among options
• THEN: Quantitative statement of mathematical modelPoisson’s Equation
–Need to analyze–Compare performance to numerical specifications
• Must recognize limits of mathematical modelNOT Math → Physical behaviorPhysical behavior first → Math
–Many students get this turned around ...
McNEILL: PCB Design in the GHz Age 26
Teaching E&M Fields
• Respond to need of engineering (not physics) students
• In GHz age, everyone needs to know E&M fields
• Problem: Surrounded by aura of mystery, magic,incomprehensibility
• Needs:1. Motivate students to take course in the first place2. Once in class, "demystify" fields
McNEILL: PCB Design in the GHz Age 27
1. Motivating Students
• Old course description–"Gauss' Law ... Ampere's Law ... Stokes' theorem ...”
• New course description–Emphasis: signal integrity in digital systems–Limit of lumped LC model: transmission line limitingpropagation velocity, system speed, signal quality
“Benefits not Features”
McNEILL: PCB Design in the GHz Age 28
2. Demystifying Fields
• Electric field demonstration relatively easy–Rub a balloon on your sweater, stick it to a wall–See lightning, spark after shuffling across carpet–Opposite charges attract–E field tells you which way charges move (forceacting on charge)
• Magnetic field–???–Try a different way to look at things...
(Acknowledgment: Paul Brokaw, Analog Devices)
McNEILL: PCB Design in the GHz Age 29
Cause and Effect: Ohm's Law
V = I R• Which is cause, which is effect?• V cause, I effect?• I cause, V effect?
I
V
-
+
R
McNEILL: PCB Design in the GHz Age 30
Chemistry vs. Electrical Engineering
V=IR
–No arrow!–V, I just "go together"–View either as cause,effect
–Whatever is best for yourpurpose
2 22 2 2H O H O+ →
Cause Effect
McNEILL: PCB Design in the GHz Age 31
You're the cause
V, I go together - Resistor doesn't care
I
V
-
+
R
I
V
-
+
R-+V I
McNEILL: PCB Design in the GHz Age 32
What causes magnetic field?
• Usual answer: current–Usually introduced in terms of current in a wirecreating an associated magnetic field
–Physics (force-based) approach
• Think about it another way ...–Better way for circuit designers–What effect is field describing?
• Look at capacitance, inductance
McNEILL: PCB Design in the GHz Age 33
Capacitance
• Represents energy stored in electric field• Capacitor voltage can’t change instantaneously• Time delay required to “build up” energy in field• C tells you how much V results from applied A-sec
I CdV
dt
VC
I dt
CC
C C
A
=
⇒ = ∫⋅
1
sec123
I V
t
I(t)
t
V(t)
McNEILL: PCB Design in the GHz Age 34
Inductance
• Represents energy stored in magnetic field• Inductor current can’t change instantaneously• Time delay required to “build up” energy in field• L tells you how much I results from applied V-sec
V LdI
dt
IL
V dt
LL
L L
V
=
⇒ = ∫⋅
1
sec123
t
V(t)
t
I(t)
I
-+V
McNEILL: PCB Design in the GHz Age 35
Different Approach to Magnetic Field
• Electric Field–Represented by capacitance–Voltage is result; cause is applied A-sec (charge)
• Magnetic Field–Represented by inductance–Current is result; cause is applied V-sec
McNEILL: PCB Design in the GHz Age 36
Units of magnetic flux density
• Magnetic flux density B
Meaning• Apply 1V for 1sec to
loop with area of 1m2
(cause)• Result is B field of
1 tesla (effect)• What about current???
teslaV
m[ ] = ⋅
sec2
I
-+V
t1 sec
1 V
V(t)
A = 1 m2
B
McNEILL: PCB Design in the GHz Age 37
What about current?
• Observed that resulting current for a given V-sec/m2
depends on material in which field lines exist• Described with permeability µ, magnetic field H:
• H gives current through Ampere’s law
H B=
1µ
I H dencl = ⋅∫ l
• Units: H must have units [A/m]⇒ Permeability µ will have units [Hy/m]
• Hint: µ will be involved in value of inductance!
McNEILL: PCB Design in the GHz Age 38
Units of electric flux density
• Electric flux density D
Meaning• Apply 1A for 1sec to
capacitor plates witharea of 1m2 (cause)
• Result is D flux of1 A-sec/m2 (effect)
• What about voltage?
coul
m
A
m2 2
= ⋅
sec
V
-
+
I
t1 sec
1 A
I(t)
D
A = 1 m2
McNEILL: PCB Design in the GHz Age 39
What about voltage?
• Observed that resulting voltage for a given A-sec/m2
depends on material in which field lines exist• Described with permittivity ε, electric field E:
• E gives voltage through path integral
E D=
1ε
V E dab ab= ⋅∫ l
• Units: E must have units [V/m]⇒ Permittivity ε will have units [F/m]
• Hint: ε will be involved in value of capacitance!
McNEILL: PCB Design in the GHz Age 40
Parallel Plate Electromagnetics
• Area = 1m2 a little large for our purposes ...• Use geometry suited to PCB design: parallel plates• Determine inductance, capacitance• Assumption: Field density negligible outside volume
enclosed by plates
l
w
h
B
McNEILL: PCB Design in the GHz Age 41
Magnetic Field / Inductance
Procedure:
• B from V-sec, areaperpendicular to field lines
• H from µ, B• Current I from H, path
integral• Inductance from definition
of L
l
w
h
B
McNEILL: PCB Design in the GHz Age 42
Magnetic Field / Inductance
B from V-sec, geometry
H from µ, B
Path integral
Inductance
l
w
h
B
t
V(t)
t
I(t)
-+V
INTEGRALPATH
VL
BV t
hL= ⋅⋅l
H B
V t
hL=
µ
= ⋅µ ⋅ ⋅
1l
I H d
w V t
hL= ⋅ = ⋅ ⋅
µ ⋅ ⋅∫ ll
I
LV dt L
h
wL=
⇒ = µ ⋅∫
1 l
McNEILL: PCB Design in the GHz Age 43
Electric Field / Capacitance
Procedure:
• D from A-sec, areaperpendicular to field lines
• E from ε, D
• Voltage V from E, pathintegral
• Capacitance fromdefinition of C
l
w
h
D
McNEILL: PCB Design in the GHz Age 44
Electric Field / Capacitance
D from A-sec, geometry
E from ε, D
Path integral
Capacitance
D
I t
wC= ⋅⋅l
H D
I t
wC=
= ⋅
⋅ ⋅1ε ε l
V E d
h I t
wC= ⋅ = ⋅ ⋅
⋅ ⋅∫ llε
VC
I dt Cw
hC=
⇒ = ⋅∫
1 ε l
l
w
h
D
I
t
I(t)
t
V(t)
INTEGRALPATH
IC
McNEILL: PCB Design in the GHz Age 45
Parallel Plate Summary
• PCB trace: Energy in both electric, magnetic fields
l l
l l l l
C
w
h= ⋅ε l
L
h
w= µ ⋅l
McNEILL: PCB Design in the GHz Age 46
Transmission Line Impedance
• Lumped L, C model (limit as segment length→0)
Z
Z Z
Z Z j L
j CZ Z
L
C
h
w
h
w
h
w= +
⇒ = =
=ω
ωµ
εµε
1 1l
l
McNEILL: PCB Design in the GHz Age 47
Qualitative interpretation
Low impedance• Wide, close to return path
(supply distribution)• High current (energy in
magnetic field)
Zh
w=
µε
w
h
High impedance• Narrow, far from return
(low power signal)• Low current (energy in
electric field)
MATERIALGEOMETRY
McNEILL: PCB Design in the GHz Age 48
“Parallel Plate” E&M summary
• Example of approach for academia:–Same concepts, different application–Responsive to needs of community
• Qualitative insight, guidance
• Quantitative predictions
McNEILL: PCB Design in the GHz Age 49
Improving Education: Industry
Create culture that supports education:
• Industry-academia relationship–Let us know what you need!
• Industry-individual relationship–Culture must allow "I don't know"–Key: Enables start of education process–What is culture’s response to "I don't know"
• Good: opportunity for education• Bad: admission of weakness
McNEILL: PCB Design in the GHz Age 50
Improving Education: Individuals
Expect to Understand!!!
• Ask Questions!!!–Ask yourself, what does that equation mean?–If you don't know, ask the professor
• If you don't understand, it's because (choose one):a) you're stupidb) it hasn't been explained to you correctly
McNEILL: PCB Design in the GHz Age 51
Professor problem
• Shocking secret:Most professors have little or no training in education
• Most professors understood material the first time–Only know (need) one explanation
• Advice for students:–Keep pushing–Connect explanation to basics–Question assumptions
McNEILL: PCB Design in the GHz Age 52
Fear Factor
• Exponential growth in material to be learned–Now teaching material to juniors that didn't exist 20years ago
• No shame in admitting "I don't know"• Don't panic
–Answer is out there somewhere• Don't reinvent wheel
–People have been looking at high speed problemsfor a long time
McNEILL: PCB Design in the GHz Age 53
Connect from analogous situations
• Repeaters for transatlantic cable (20th century):
• Repeaters for on-chip clock distribution (21st century):
≈ 60km
fmax ≈ 200kHz
≈ 1mm
fmax ≈ 2GHz
Ismail, Friedman, Neves, “Repeater Insertion in Tree Structured Inductive Interconnect, IEEE TCAS, May, 2001
McNEILL: PCB Design in the GHz Age 54
Conclusion
We've all got work to do:
• Academia:–More learner centered, meet needs of students
• Industry:–Create, maintain culture that supports education
• Individual:–Expect to understand, question assumptions