signal integrity: problems and solutions - keysight bogatin 2000 slide -1 myths signal integrity:...
TRANSCRIPT
Eric Bogatin 2000
Slide -1
www.bogatinenterprises.com
MYTHS
Signal Integrity:Problems and Solutions
Dr. Eric BogatinPresident
Bogatin Enterprises
www.BogatinEnterprises.com(copies of the presentation are available for download on the web site)
Presented at Lockheed, Sunnyvale, CA, March 1, 2000
Eric Bogatin 2000
Slide -2
www.bogatinenterprises.com
MYTHS Overview
• What is Signal Integrity?
• Why is it growing in importance?
• What can you do about it?
Eric Bogatin 2000
Slide -3
www.bogatinenterprises.com
MYTHS
Signal Integrity and Interconnect Design
How the electrical properties of the interconnects screw up the beautiful,
pristine signals from the chips
Eric Bogatin 2000
Slide -4
www.bogatinenterprises.com
MYTHS
The Confusing Mix of Signal Integrity Problems
LOSSY LINES
CROSSTALK
PARASITICS
EMI/EMC
GROUND BOUNCE
INDUCTANCE
EMISSIONS
TRANSMISSION LINES DELTA I NOISE
IR DROP
ATTENUATION
RC DELAY
POWER AND
GROUND DISTRIBUTION
CRITICAL NET
SIGNAL INTEGRITY
SKIN DEPTH
RETURN CURRENT PATHSTUB LENGTHS
TERMINATIONSCAPACITANCE
GAPS IN PLANES
REFLECTIONS
RINGING
LINE DELAY
UNDERSHOOT, OVERSHOOT DISPERSION
LOADED LINES
SUSCEPTABILITY
IMPEDANCE DISCONTINUITIES
NON-MONOTONIC EDGES
Eric Bogatin 2000
Slide -5
www.bogatinenterprises.com
MYTHS The Four High Speed Problems
1. Signal quality of one net: reflections and distortions from impedance discontinuities in the signal or return path
2. Cross talk between multiple nets: with ideal return paths, and without (SSO)
3. Rail collapse in the power and ground distribution network
4. EMI from a component or the system
Eric Bogatin 2000
Slide -6
www.bogatinenterprises.com
MYTHS
Signal Quality on One Net:Distorted by the Interconnect
Initial output signal
Signal distorted by interconnect
Simulated with Hyperlynx
Eric Bogatin 2000
Slide -7
www.bogatinenterprises.com
MYTHS
Cross Talk Between Two Adjacent Conductors- Ideal Return Path
Near end
Far end
50ΩΩ
(HP 83480 High speed scope and TDR)
Active line
far end
Near end
The far end noise is ~ 10x larger than the near end noise
rise time ~ 100 psec, TD ~ 1 nsec
Eric Bogatin 2000
Slide -8
www.bogatinenterprises.com
MYTHS
On Chip
V SS
VCC
GND
15836© 1991 Integrated Circuit Engineering Corporation
L Bonding
L BondingPower
common lead
inductance
Conceptual Origin of SSO Noise
Icharge
Idischarge
Switching lines
Quiet data line
Eric Bogatin 2000
Slide -9
www.bogatinenterprises.com
MYTHS
Simple Example of Rail Collapse
100 nF
Rail collapse: ∆∆V ~ - dI/dt
Source: National Semiconductor
CdecouplingTo
regulator
Current On Current Off
Vdd nominalVdd rail collapse
Eric Bogatin 2000
Slide -10
www.bogatinenterprises.com
MYTHS
Radiated Emissions and Power and Ground Routing
Eric Bogatin 2000
Slide -11
www.bogatinenterprises.com
MYTHS
Two Classes of High Speed Problems
• Timing: setup, hold, propagation delay, skewü Scales with decreasing clock period
• Electrical Noise: signal integrity and EMIü Scales with decreasing rise time
dIdt
dVdt
f, f2
Eric Bogatin 2000
Slide -12
www.bogatinenterprises.com
MYTHS
On Chip
V SS
VCC
GND
15836© 1991 Integrated Circuit Engineering Corporation
L Bonding
L BondingPower
common lead
inductance
“…it’s the rise time, …”
Icharge
Idischarge Switching data lines
Quiet data line
N = number of switching leads per ground leadsL = lead inductance or lead lengthττ = rise time
SSO noise ~N x L
ττ
Eric Bogatin 2000
Slide -13
www.bogatinenterprises.com
MYTHS
Shorter Delays Mean Shorter Clock Periods, Higher Clock Frequencies
Digital Clock Frequencies are Increasing: doubling every 2 years!
1
10
100
1000
10000
1970 1975 1980 1985 1990 1995 2000
Introduction Year
Clo
ck F
req
uen
cy (
MH
z)
Clock frequency of Intel Processors
High speed usually refers to increasing clock frequency
Eric Bogatin 2000
Slide -14
www.bogatinenterprises.com
MYTHS Increase in Clock Frequencies
0
500
1000
1500
2000
2500
3000
3500
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014
Year
Clo
ck F
req
uen
cy (
MH
z)
on-chip
on-board
Source: SIA Roadmap
Eric Bogatin 2000
Slide -15
www.bogatinenterprises.com
MYTHS
Rise Times Are Loosely Related to Clock Frequency
0.01
0.1
1
10
100
1 10 100 1,000 10,000
Clock Frequency (MHz)
Ap
pro
xim
ate
Ris
e T
ime
(nse
c)
clockF1
101
~τWhat is the consequence of higher speed?
10 nsec period
1 nsec rise time
Eric Bogatin 2000
Slide -16
www.bogatinenterprises.com
MYTHS
The Driving Force Fueling the Electronics Revolution:
Gate Length Feature Size Reduction
50% reduction every 4 years
Eric Bogatin 2000
Slide -17
www.bogatinenterprises.com
MYTHS
Transistors Switch Faster As Channel Length Shrinks
Shorter channel length means:->> shorter delay->> shorter rise time
in out
What can happen to the clock period and clock frequency?
Eric Bogatin 2000
Slide -18
www.bogatinenterprises.com
MYTHS Situation Analysis
• Clock frequency will get faster• Rise times for everyevery chip will get shorter• SI problems will be more significant• Design cycle times will be decreasing
Conclusion:Getting new products to market on time will be harder.
Solution: A new design methodology is needed.
Eric Bogatin 2000
Slide -19
www.bogatinenterprises.com
MYTHS The Old Design Strategy
Guess a design
Hope it works
Build it
Test it
Try to Fix it
Ship it
Eric Bogatin 2000
Slide -20
www.bogatinenterprises.com
MYTHS
Details of the Three Design Approaches
Source: G. Doyle, Mentor Graphics
Design by correcting
Design by virtual iteration
Correct by designThe earlier in the design cycle problems can be identified and solved, the lower the development cost and the faster time to market.
Eric Bogatin 2000
Slide -21
www.bogatinenterprises.com
MYTHS
Two Critical Processes for Virtual Design and Test
Modeling: Translating the physical world into an equivalent electrical circuit model (Schematic)
Simulation: Predicting voltage/current waveforms based on the circuit behavior
Zo,
τ
D Zo,
τ
D
Lpower
Lgnd
Clk1
Lpin
Cpin
LpinLconnLconn
Cconn Cconn
Cpin
Lpower
Lgnd
Clk1
Gate1
Gate2
PCB #1 Backplane PCB #2
Zo,
τ
D
V1
PULSE
R1 50 L1 1U
C1
30P
Q 2
QN3904
Q10
QN3906
R4
680
R2
5K
V3 10
V2 10
X1 WIRER3 10
C2
7P
V(10)
VLOADV(7) VEMITTER
V(3)
VOUT
2 1 3
4
7
8
9
6 10
Eric Bogatin 2000
Slide -22
www.bogatinenterprises.com
MYTHS
1
6
11
16
21
26
M01:i011
M09:i091
M16:i161
M24a:i24a1
0.0
0.5
1.0
1.5
2.0
Ind
uct
ance
(n
H)
• Calculations: (03, 06)ü Rules of thumbü Analytic approximationü Parasitic extraction numerical tools: field solvers
• Measurements: (06)ü Impedance analyzer (LCZ)ü Network Analyzer (NA)ü Time Domain Reflectometer (TDR)
Where do Models Come From?
Courtesy of TDA Systems
Eric Bogatin 2000
Slide -23
www.bogatinenterprises.com
MYTHS
Two Tools for Simulating Circuits
• SPICE: Simulation Program with Integrated Circuit Emphasisü PSPICE from OrCAD/Cadenceü IsSPICE from Intusoftü Advanced Design System (ADS) from HP Eesofü Maxwell SPICE from Ansoftü Micro-CAP from Spectrumü HSPICE from Avant!
• IBIS based simulators: Input/output Buffer Interface Specification ü Hyperlynx (Pads)ü Veribest/Mentor Graphicsü Zukan Redacü Viewlogicü Interconnectix (Mentor Graphics)
Eric Bogatin 2000
Slide -24
www.bogatinenterprises.com
MYTHS Design Principles for Good SI
Noise Categories Design Principles
Signal Quality Signals should see the same impedance through all interconnects
Cross talk Keep spacing of traces greater than a minimum value, minimize mutual inductance of non ideal returns
Rail Collapse Minimize the impedance of the power and ground path
EMI Minimize bandwidth, minimize ground impedance and shield
When are you done? How much reduction is enough?
Eric Bogatin 2000
Slide -25
www.bogatinenterprises.com
MYTHS
time
money
risk
Cost factors:
…just follow these RULES
PerformancePerformance(meet specs)(meet specs)
Eric Bogatin 2000
Slide -26
www.bogatinenterprises.com
MYTHS
Design Tradeoffs Are Negotiated With a Budget
• Total voltage swing is 3.3v• Within 500 mV, all the noise sources must be accounted for:
*
*dynamic effects important
An example:
•• In hi speed systems, keeping within the noise budget is HARD!In hi speed systems, keeping within the noise budget is HARD!•• The more accurately you can predict performance, the less marginThe more accurately you can predict performance, the less margin needed and the higher the needed and the higher the
performance performance
Noise Source Allocated BudgetRinging/reflections 100mVDiscontinuities 40mVCross talk 90mVSSO noise 120mVRail collapse 100mVTotal* 450mVMargin ~50mV
Discontinuities9%
Rail collapse22%
SSO noise27% Cross talk
20%
Ringing22%
Eric Bogatin 2000
Slide -27
www.bogatinenterprises.com
MYTHS
The Most Important General Design Principles
1. Slow down edges
2. Minimize the length of all interconnects
3. Use low dielectric constant materials for signal layers
4. Use controlled impedance lines and terminate
5. Minimize loop mutual inductances between signal lines
6. Use continuous, closely spaced, adjacent power and ground planes
Eric Bogatin 2000
Slide -28
www.bogatinenterprises.com
MYTHS
#1 solution: slow down the edges
50 psec
150 mils spacing
50 Ohm line
2 short stubs (capacitive discontinuity)Top view
Longer the rise time, smaller the impact, or,
the shorter the discontinuity, the smaller the impact
Eric Bogatin 2000
Slide -29
www.bogatinenterprises.com
MYTHS Minimize Bandwidth
AVX Z chip: integrated RC, with low stray C
Spread Spectrum Clock Generator (SSCG)
At 2 GHz At 2.3 GHz
Figure 28. Data from Ansoft HFSS showing the field distribution on andoff resonance for a 208 lead QFP, excited at one lead.
Avoid resonance and clock harmonics
Eric Bogatin 2000
Slide -30
www.bogatinenterprises.com
MYTHS
#2 solution:shorter is better
• Reflections:
• Cross talk
• Rail collapse
• EMINear end
Mutual C, mutual L, scale with lengthSeries L scales with lengthRadiated emission scales with length of current path
Eric Bogatin 2000
Slide -31
www.bogatinenterprises.com
MYTHS
Terminations will Minimize Reflected Noise from the Ends
Series R terminate
RC terminate at far end, changing C
Source: Analog Devices
Eric Bogatin 2000
Slide -32
www.bogatinenterprises.com
MYTHS Avoid Stubs and Branches
(for 0.5 nsec edges, stub length < 0.5 inches)
branches
daisy chain
Eric Bogatin 2000
Slide -33
www.bogatinenterprises.com
MYTHS
Thin Power and Ground Layers Reduce Switching Noise
Conventional, 10 mil thick spacing, 2 plane pairs
Thin layer, 2 mil thick, 4 plane pairs“A Low-Cost Technique for Reducing the
Simultaneous Switching Noise in Sub-Board Packaging Configurations”, Koike and Kaizu, IEEE Trans CPMT part B vol 21(4) Nov 1998 p. 428
Small daughtercard
Eric Bogatin 2000
Slide -34
www.bogatinenterprises.com
MYTHS Reduced Switching Noise
Improves effectiveness of the decoupling caps
Reduces SSO noise
Eric Bogatin 2000
Slide -35
www.bogatinenterprises.com
MYTHS
Reducing Emissions: Low Impedance Power and Ground Layers by
Thinner Dielectric
Eric Bogatin 2000
Slide -36
www.bogatinenterprises.com
MYTHS Avoid Splits in Return Path
Archambeault, Bruce; “Proper design of intentional splits in the ground reference plane of PC Boards to minimize emissions from I/O wires and cables”, Proc. 1998 IEEE conf on EMC, p. 768
with split
no split
Avoid all splits in the return path!Avoid all splits in the return path!
Eric Bogatin 2000
Slide -37
www.bogatinenterprises.com
MYTHS Unintentional Splits
Figure 9. How a via field for a connector can create a gap. By decreasingthe clearance hole diameter in the ground plane, a continuous returnpath can be provided.
Figure 10. Data from [10]. Left is the emission from a board with gapsunder via fields- failing the Class A test. Right: the exact same board, butwith smaller clearance holes and no gaps under traces- passing Class Atest.
Decreasing size of clearance holes reduced radiated emissions
Eric Bogatin 2000
Slide -38
www.bogatinenterprises.com
MYTHS The Design Strategy
1. Use design guidelines as design guidelines to shoot for
2. Estimate the magnitude of each effect and the benefit from a design or technology solution
3. Verify the models and simulations based on measurements of test vehicles and previous designs
4. Evaluate cost/performance trade offs
5. Keep optimizing until the noise budget is met
6. The earlier in the design cycle correct design decisions can be made, the shorter time to market and lower the development cost
Eric Bogatin 2000
Slide -39
www.bogatinenterprises.com
MYTHS
SI Problems Apply Across ALL Interconnects
Courtesy of ICE
BOLData Corp
Eric Bogatin 2000
Slide -40
www.bogatinenterprises.com
MYTHS
“There are two kinds of design engineers, those that have signal
integrity problems, and those that will”
Good Luck!