google zhiping yang, ph.d
TRANSCRIPT
The myth about ideal ground
Zhiping Yang, Ph.D.Google
Quiz
Which package model is the best at high frequency?
L_PWR
L_GND
L_sig1
L_sig2
L_PWR
L_GND
L_sig1
L_sig2
L_PWR
L_sig1
L_sig2
M12M12 M23
M13
M23
M13
M24
M14M34
A B C
Content
● What are the myths about ground
● History of ground in electrical design
● Terms related to ground
● A package example with signal and ground nets only
● A package example with ground, signal and power nets
● S-parameter and W-element models in high-speed modeling and simulation
● Conclusions
What are myths about ground
● Myth #1: Does ideal ground really exist? If not, why do we use it in modeling and simulation?
● Myth #2: Should we put partial inductance and resistance on the ground net? (i.e. is ground bounce real?)
● Myth #3: Should I have a different reference node for each port of W-element or S-parameter model?
History of ground in electrical design
● Ground was first used by telegraph system to reduce wiring cost. (Discovered by German scientist Carl August Steinheil in 1836-1837)
●
History of ground in electrical design
● Later, power engineers used the ground as return path in Single Wire Earth Return (SWER) electrical distribution system
● The ground is connected to protective earth (PE) or Neutral (N) conductor of modern power systems for prevention of electrical shock (Safety) and fault detection and protection
● The ground of circuit boards and IC chips are typically connected to the metal box and ground of AC power systems
History of ground in electrical design (continued)
● The ideal ground concept (zero voltage and infinite current sink capability) was introduced to analog and digital circuits to simplify circuit schematics and analysis.
● The ideal ground node was used in SPICE-type simulators to solve KCL and KVL more effectively.
● At high frequency, the concept of local ground/reference was used to get more precisely and meaningful results
One of the unwanted misimpressions from ideal ground node: The voltage signal can travel from driver to receiver on a single wire.
A true story about ground
Daddy, what are they for?
This is…, so you'll be grounded...
Cisco Lab
Grounded! GROUNDED? What did
I do wrong?!!
Terms related to ground
● Global Ground, Ideal Ground and Virtual Ground
● Local Ground or Local reference
● Self Loop Inductance, Mutual Loop Inductance, Partial Self Inductance, Partial Mutual Inductance, Partial Loop Self Inductance, and Partial Loop Mutual inductance.
● Ground Bounce, Power Bounce and Voltage Bounce.
“Ground is a place where potatoes and carrots thrive”—by Bruce R. Archambeault, PCB Design for Real-World EMC Control
A package example with signal and ground nets only
Die side
PCB side
i2
i1
L1
W1
a1b1
b2
d1c1
a2
c2 d2Cutout on ground plane
W1
H1T2
S1
W2
A valid circuit model at low frequency
Good things about this model:1. It is valid at very low frequency at which current is evenly distributed
across the ground plane and c1 & c2 ( d1&d2) have the same voltage potential.
2. You can actually probe the voltage across any two points in this circuit and it is meaningful and could match the measured data very well, so ground bounce (voltage difference between c1/c2 and d1/d2) is well defined.
Die side PCB side
R1a1 b1
R2a2 b2
i2
c1 d1i1+i2
c2 d2Rgnd
L2
L1
Lgnd
M1
M2
M3
i1
What could change at high frequency?
Die side
PCB side
i2
i1
L1
W1
a1b1
b2
d1c1
a2
c2 d2Cutout on ground plane
W1
W2
H1T2
S1
Is previous low-frequency model still valid at high frequency?
Die side PCB side
R1a1 b1
R2a2 b2
i2
c1 d1i1+i2
c2 d2Rgnd
L2
L1
Lgnd
M1
M2
M3
i1
What could be wrong with this model?1. It is clearly shown that the return current for conductor 1 and 2 are not the same.
How can Rgnd and Lgnd be well defined?2. How to separate the L1, Lgnd and M1 from the physical loop inductance? No
unique definition for L1, Lgnd and M1.3. At high-frequency, when wavelength is close to the structure size, (such as L in the
previous figure), the voltage is not well defined between die ground and PCB ground. Local reference voltage definition is the only meaningful voltage at high frequency.
A working circuit model at high frequency
Compared with the previous model, the above is a much better and accurate model. The only drawback is too many components (especially partial and mutual inductances) and these inductances are not well defined. Since we only care about the voltage with local reference, this model can be further simplified by model reduction.
Die side PCB side
R1a1 b1
R2a2 b2i2
c1 d1
i1
c2 d2
Rgnd1
L2
L1Lgnd1 M1
M2Rgnd2
Lgnd2 M3
i2
M4R12
A Simplified and better model
Die side PCB side
R1+Rgnd1
a1 b1
R2+Rgnd2
a2 b2
i2
c1 d1i1
c2 d2
L2+Lgnd2-2*M2
L1+Lgnd1-2*M2
i1
M3+M4
i2
R12
This model is much simpler and it is accurate from DC to very high-frequency. All components in this circuit are well defined and have unique physical meanings.
The ground nodes c1/c2/d1/d2 could be shorted together (by doing so, it loses the mapping between physical locations and circuit nodes.
A package example with power, signal and ground nets
Die side
PCB side
i2
i1
L1
W1
a1b1
b2
d1c1 q2
c2 d2
Cutout on ground plane
q1
p1 a2 p2
W2
H1T2
S1Ground
VDD
VDDQ
Which package model is better?
Die side PCB side
i_s1
R1’a1 b1
R2’a2 b2
i_s2
c1c2
p2i_vdd
q1 q2
L2’
L1’
Lp’
M1’
M2’
Rq’Lq’
M3’
M4’
Rp’p1
d1d2
i_vddqM5’
M6’
i_gnd
Die side PCB side
R1
a1 b1
R2
a2 b2i_s2
c1
c2
p2i_vdd
q1 q2
Rgnd
L2
L1
Lp
M1
M2
i_s1
Rq
LqM3
M4
Rpp1
d1d2
Lgnd
i_vddq
M5
M8
M6
M7M9
M10
i_gnd
a. Package model with ground parasitic
b. Package model without ground parasitic
R12
R23
R34
R13
R24R14
S-parameter and W-element model in high-speed modeling and simulation
i2
i1
L1
W1
a1b1
b2
d1c1
a2
c2 d2Cutout on ground plane
Port #1
Port #2
Port #3
Port #4
S4P S-parameter file
Ref
a1
a2
b1
b2
a. S-parameter file connection case #1
S4P S-parameter file
a1c1a2c2
b1d1
b2d2
b. S-parameter file connection case #2
More complicated port settings
i2
i1
L1
W1
a1b1
b2
d1c1
a2
c2 d2Cutout on ground plane
Port #1
Port #2
Port #3 is a differential-mode portPort #4 is a common-mod port
S4P S-parameter file
Ref
a1
a2
dif
com
a. S-parameter file connection case #1
S4P S-parameter file
a1c1a2c2
difref1mmcomref2
b. S-parameter file connection case #2
Suggestions for the application of S-parameter model
● Do not add circuit components between positive port terminals if their negative terminals are not defined on the same surface and conductor(s).
● It is better to have common reference conductor(s) for all ports located on the same surface.
Quiz
Which package model is the best at high frequency?
L_PWR
L_GND
L_sig1
L_sig2
L_PWR
L_GND
L_sig1
L_sig2
L_PWR
L_sig1
L_sig2
M12M12 M23
M13
M23
M13
M24
M14M34
A B C
Conclusions● Ground bounce and power bounce are only valid terms
at low frequency.● Voltage bounce is more precise and should be used at
high frequency.● Partial loop inductance and partial mutual inductance
are well defined and have unique values.● Package model without parasitic in ground net is not
only valid, but also more computationally effective and accurate.
● S-parameter based model should be connected with other circuit elements in ways which are consistent with the port definition when it was extracted.
Q&A
?