You are on page 1of 4

978-1-4673-1036-9/12/$31.

00 2012 IEEE 523 13th Int'l Symposium on Quality Electronic Design


Maintaining Power Integrity by Damping the Cavity-Mode Anti-Resonances Peaks
on a Power Plane by Particle Swarm Optimization

Jai Narayan Tripathi
1
, Raj Kumar Nagpal
2
, Nitin Kumar Chhabra
2,
Rakesh Malik
2
and Jayanta Mukherjee
1
1
Department of EE, Indian Institute of Technology Bombay, Mumbai, INDIA
2
STMicroelectronics Pvt. Ltd., Greater Noida, INDIA
1
E-mail: {jai,jayanta}@ee.iitb.ac.in
2
E-mail: {rajkumar.nagpal,nitin.chhabra,rakesh.malik}@st.com

Abstract
To maintain Power Integrity in a high speed system, an
effective methodology for suppressing the cavity-mode anti-
resonances peaks is presented. The optimum values and the
optimal positions of the decoupling capacitors are found
using Particle Swarm Optimization, which leads to optimum
impedance of power plane loaded with decoupling capacitors.
Optimum number of capacitors and their values, by which
impedance of loaded board is matched below the target
impedance of the system, are found.
Keywords
Power Integrity, Power Plane, Cavity-Mode Anti-
Resonances, Decoupling Capacitor, Target Impedance,
Particle Swarm Optimization.
1. Introduction
Power Integrity (PI) refers to the uninterrupted, sufficient
and efficient distribution of power within a system. Power
Integrity is becoming the major issue as the operational
frequencies of the integrated circuits are increasing up to
GHzs [1]. If not maintained properly, Power Integrity may
affect the functionality of a high speed system. PI is
associated with Power Delivery Networks (PDNs). PDN
should be designed in such a way that PI should be taken care
at a range of frequencies which may affect the system. PDN
mainly consists of on-chip power delivery system and off-
chip power delivery system. Power planes are used for off-
chip power delivery as a power supplier from Voltage
Regulator Module (VRM) and bulk capacitor to the I/O of
integrated circuits. Power planes behave as cavity resonators
and thus affect the power delivery of the system. There are
both capacitive and inductive behaviors of power planes.
These opposite behaviors form the resonance and anti-
resonance patterns depending upon the physical and electrical
properties of the cavity [2]. Decoupling capacitors are used to
nullify the effect of the anti-resonance peaks which provide
much higher impedance at certain frequencies. The position
of the capacitors used for such decoupling strategy, is also
important. This paper takes into account the placement of
decoupling network, to maintain the uninterrupted power
supply.
For Electronic Design Automation (EDA), statistical
methods, stochastic methods or deterministic optimization
methods are used - such as Design Of Experiments (DOE)
[3], Genetic Algorithm (GA) [4], Geometric Programming
[5] etc. Stochastic methods are useful when the behavior of
system is not easy to determine. In this paper we have used

one such stochastic method called Particle Swarm
Optimization (PSO) for finding the optimum values and
positions for decoupling capacitors on a power plane of a
PDN. Finding the position and value of decoupling capacitors
on a power plane by GA is described earlier by Sungtek
Kahng [4]. In [4], the cavity modes are suppressed by
decoupling capacitors but target impedance is not taken into
account. Target Impedance of a system is concerned with the
maximum allowable ripple in the power supply rail when the
maximum transient current is defined.

Z
tugct
=
maxImum aIIowcd rIppIc oI suppIy voItagc
maxImum currcnt
(1)

The power pins of the chip must see target impedance at the
power plane. The impedance of the power delivery system is
expected to be lesser than this, in order to maintain the PI of
the system. Here, we present a general methodology to meet
the target impedance requirements by placing decoupling
capacitors on optimal positions.
2. Theory
2.1. Cavity Model for Power Planes
Power planes are the simplest structures used as a power
bus. At lower frequencies, these can be analyzed as lumped
components but at higher frequencies distributed modeling is
necessary. These can be considered as rectangular cavities for
easier computations, compared to the full wave numerical
modeling techniques [6]-[9].

Figure 1 : Power Plane Cavity Model

Tripathi et al, Maintaining Power Integrity

Z
Ld
(, X
],

]
) =

m=0

n=0

y
mn
.c
mn
(X
s,

s
).c
mn
(X
],

]
).
W
z
(W
x
W
j
)
sn
Q
+]_so-
k
xm
2
+ k
jn
2
n
_+
y
mn
.W
z
W
x
W
j
c
mn
2
N
u
u=1
(X
Lu
,
Lu
).
Lu
(2)

If the cavity model is taken into account for computation,
the impedance of the power plane can be expressed as a
double infinite sum of the terms corresponding to the cavity
modes. This is shown in equation 2. The impedance Z
Ld
(f, X
f

,Y
f
) of a power plane (of the size W
x
W
y
W
z
) observed at
(X
f
, Y
f
) which is point M in fig. 2, fed by current I
s
at source
point (X
s
, Y
s
), loaded by N
u
loads at positions (X
L
, Y
L
), can be
given by equation 2 [4]. Where
c
mn
(X
,

) =
cos(k
xm
X

). cos(k
n

). sinc
k
xm
P
xi
2
. sinc
k
jn
P
ji
2
(3)
k
xm
=
m
w
x
, k
n
=
n
w
j
, = 2n (4)
= jtono +2p
o
kw
z
2
[
-1
(5)

L

= c
mn
2
N
u
u=1
(X
Lu
,
Lu
).
Lu
(6)

Lu
= jR
Lu
+ ] [I
Lu
-
1
oC
Lu
[
-1
(7)

mn
is 1 if both m and n are 0, is 4 if m 0 as well as n 0
and is 2 if any of them is 0. tan , , , f and P
i
are loss
tangent, permittivity, permeability, frequency and ports
width respectively. In equation 2, j is imaginary number
-1. Y
Lu
corresponds to the admittance of u
th
load and Y
L
is
sum of Y
Lu
for all the loads. In a bare board, since there is no
load the term
L

vanishes. Figure 2 shows the impedance


profile up to 1 GHz, for the board without any load, having
the dimension 200 mm 150 mm 0.17 mm, source placed at
point (0,75mm) and impedance observed from the point
(0,75mm). The port width and length, both are taken 0.254
mm. The values of tan ,
r
, and
r
were taken 0.02, 4.1 and 1.
In eq. 5, k is the conductivity of the metal which is 59.6 10
6
s/m.


2.1. Particle Swarm Optimization
Particle Swarm Optimization (PSO) is a meta-heuristic
optimization technique belonging to the group of algorithms
inspired by the nature. Introduced by Kennedy and Eberhart
in 1995, PSO is inspired by the movement of fishes while
schooling and the same of the birds while flocking. The
fishes or birds follow the group behavior and the collective
intelligence is used for the movement of the entire swarm. In
PSO, there are P particles generated or defined randomly in a
design space. Each particle is assigned a position and an
initial velocity. Each particle claims to be a solution of the
problem within the search space, tending to attain optimum
position. After generating particles, the fitness of all the
particles is calculated. Fitness is the output parameter or
objective function of the optimization problem. Particles
move towards the fittest particle and by this process they
reach to the optimal solution. After calculating fitness of all
the particles, the one with the best fitness is defined as
globally best particle gbest. The movement of all the
particles is decided by the surrounding particles. After
calculating the fitness of all the particles, particles are
assigned new velocities and positions according to the
following equations [10][11] :
u

(t +t)
= (t)u

(t) + p
1
r
1
(x
I
-x

) +p
2
r
2
(x
g
-x

) (8)
x

(t +t) = x

(t) + +u

(t +t) t (9)
(t) = (

-
]
)
(t
mcx
-t)
t
mcx
+
]
(10)
where x
i
is the position of a particle and v
i
is the velocity, i is
the particle index, t is the current iteration number, t is
equal to 1 (used to jump to next iteration). r
1
and r
2
are
random numbers uniformly distributed in the range [0, 1].
Each particle has the memory to store its best position till
now, denoted by x
l
. The parameter is inertia, and p
1
and
p
2
are the acceleration coefficients. The x
g
represents the best
position attained by globally best particle.
i
and
f

represent the initial and final values of respectively, and
t
max
is the maximum number of iterations. The second term of
eq. 8 is called cognitive component while the last term is
called social component. According to Kennedy (2006),
new position of a particle is decided by its current position,
persistence and social-influence [12][13]. Thus in each
iteration, particle considers the impact of the collective
behavior. After the iteration, each particle moves to a new
position and attains a new velocity, according to eq. 8 and eq.
9. is updated linearly as given in eq. 10.
3. Optimization
For applying PSO, the parameters or variables taken were
the r, l, c values of the capacitors and their locations on
power plane represented by co-ordinates x, y. Thus, the
positions and the velocities of the particles are five
dimensional each. The problem was defined as continuous
PSO problem where the ranges of values of capacitor
parameter were taken from the options available in the
market [14]. The ranges are given in table I.

Table I: Ranges for parameters of decoupling capacitors

There are two cases discussed using PSO, one with one
decoupling capacitor and another one with two decoupling
capacitors.
Parameter Lower Bound Upper Bound
r () 0.1 5
l (nH) 0.01 12
c (nF) 0.001 18
x (mm) 0 200
y (mm) 0 150
Tripathi et al, Maintaining Power Integrity


Figure 2: Power Plane Impedance Profile by Cavity Model
3.1. One Decap Optimization
From the impedance profile of unloaded power plane
observed from the point (X
s
, 0) (as shown in fig.2), the first
anti-resonance was observed at 365 MHz. At this frequency
the impedance was maximum in the given frequency range of
interest (up to 1 GHz). According to the parameters given in
table I, 75 random particles were defined for 50 iterations.
The values of
i
,
f,
p
1
and p
2
were taken as 0.9, 0.4, 1.49
and 1.49 respectively [15]. The particles converged and the
solution was [0.1, 0.037 nH, 10 nF, 200mm,150mm],
resulting the maximum impedance (100 MHz to 1 GHz) of
0.9480 . The impedance profile of the globally best particle
(final solution) is shown in fig. 3. It can be seen that it has
damped the anti-resonance peaks.


Figure 3: Impedance profile of board with one decoupling
capacitor of optimal value and loaded at optimal position
3.2. Two Decap Optimization
The same optimization problem mentioned above, when
solved using two capacitors, required ten dimensional
particles (5 parameters for each capacitor). For 75 particles
and 50 iterations, the solution was [0.1, 0.01 nH, 11.194
nF, 200mm, 150mm] for one capacitor and [0.1, 9.956 nH,
3.587 nF, 188.63mm, 66.12mm], for another one, resulting
the maximum impedance (up to 1 GHz) of 0.7352 as
shown in fig.4. Fig. 4 shows the damping of peaks with both
the cases i.e. using one and two decoupling capacitors.
Figure 4: Impedance profile of board with two decoupling
capacitors of optimal values and loaded at optimal positions
4. Solving Power Integrity Problem
As shown in the previous section, the impedance of the
system was lowered by placing capacitors. But even if two
capacitors, it doesnt match to the target impedance. In our
system the voltage is 3.3 V with the tolerance of device equal
to 8%. The device draws 0.5 A current, thus target impedance
needed is 0.528 . So a computer program was written to
find the optimal number of capacitors as well, in addition to
their values and positions. The minimum number of
decoupling capacitors and their corresponding values, needed
to achieve this target impedance were found by PSO. The
pseudo-code for the search process is given following:
4.1. Pseudo-Code:
1) Define P particles, their respective positions x (i, j, 1) and
velocities v (i, j, t + 1); where i {1, 2, , P} and j {1, 2,
3, 4, 5}, within the lower and upper bounds.
2) Calculate the maximum impedance peak max_imp(i, 1)
corresponding to all the i particles at time t = 1 (which is the
impedance of the board loaded with N = 1 decoupling
capacitor).
3) Define all the local particles as local best particles lbest
(i, j, t) at t = 1, and find the global best particle gbest (i, j).
4) Loop 1 : for t = 2:T
5) Update inertia, velocities and positions
(t) = (

-
]
)
(t
mcx
-t)
t
mcx
+
]
(11)
u

(i, ], t +1)
= (t)u

(i, ], t) + p
1
r
1
(|hext(i, ], t) -x(i, ], t)

) +
p
2
r
2
(ghext(i, ], t) -x(i, ], t)

) (12)
x(i, ], t +1) = x(i, ], t) + u

(i, ], t +1) (13)
6) Limit the positions and velocities within the lower and
upper bounds.
7) Update lbest (i, j, t) and gbest (i, j) accordingly.
8) End : Loop 1
9) Check if {imp_gbest (t) imp_target} then N = N + 1 :
Loop 1
10) Final Solution = gbest (t)
5. Results
The number of decoupling capacitors N needed to meet the
target impedance were 5. Their values and positions were
also found by PSO. Table II shows the r, l, c values and
Tripathi et al, Maintaining Power Integrity

positions of the capacitors needed to achieve the target
impedance.

Table II: Parameters and positions for optimum decoupling
capacitors
The impedance profile of the plane after loading these
capacitors at their best positions is shown in fig. 5. It can be
seen that the maximum impedance in the frequency range of
interest, is less than the target impedance. The maximum
impedance after loading by these 5 caps, is 0.5088 .

Figure 5. Meeting the target impedance by optimal number
of decoupling capacitors.
6. Conclusion
PSO is proved to be useful for designing Power Delivery
Network, in order to maintain Power Integrity. The ease of
implementation of PSO was the motivation of using it. The
optimal number of decoupling capacitors, their values and
their positions were found by PSO. A pseudo-code is given to
solve the power plane target impedance problem, which can
be used for other geometries also.

7. References
[1] Raj Kumar Nagpal, Rakesh Malik, Jai Narayan Tripathi,
Signal Integrity and Power Integrity Methodology for
Robust Analysis of On-the- Board System for High Speed
Serial Links, 12
th
Euromicro Conference on Digital System
Design, (DSD 2009), 27-29 Aug. 2009, Patras, Greece.
[2] M Swaminathan and A. Ege Engin, Power Integrity
Modeling and Design for Semiconductors and Systems,
Prentice Hall, 2008.
[3] Jai Narayan Tripathi, Raj Kumar Nagpal, Rakesh Malik,
Robust Optimization of Serial Link System for Signal
Integrity and Power Integrity, 1
st
IEEE International
Conference on Networked Embedded Systems for Enterprise
Applications (IEEE NESEA), Nov. 25-26 2010, Suzhou,
China.
[4] Sungtek Kahng, GA-Optimized Decoupling Capacitors
Damping Power Bus Cavity-Mode Resonances, IEEE
MWCL, Vol.16, No.6, June 2006.
[5] D. M. Colleram et al, Optimization of Phase-Locked
Loop Circuits via Geometric Programming, IEEE CICC, pp.
377-380, Sept. 21-24, 2003. Suzhou, China.
[6] G. T. Lei et al, High-frequency characterization of
power/ground-plane structures, IEEE Transactions on
Microwave Theory & Technology, vol. 47, pp. 562-569, May
1999.
[7] O. Wada et al, Local decoupling effects of decoupling
capacitors on a multilayer PCB: Fast analysis by a closed
form expression, IEICE Tech. Rep. EMCJ2000-115, vol.
100, Dec. 2000.
[8] X. Yang , Z. Li, and J. Mao, Analysis of the bounces on
the power/ground plane structures by planar circuit model
and APA-E algorithm, IEEE Transactions on Advanced
Packaging, vol. 24, pp. 184-190, May 2001.
[9] S. Chun et al, Modeling of simultaneous switching noise
in high speed systems, IEEE Transaction on Advanced
Packaging, vol. 24, pp. 132-142, May 2001.
[10] Jai Narayan Tripathi, Designing, Optimization and
Modeling of Analog/RF Circuits by Design of Experiments,
Ph.D. Forum, 19
th
IEEE/IFIP

International Conference on
Very Large System Integration, VLSI-SoC, pp. 457-460, Oct.
2-5, 2011, Hong Kong, China.
[11] Mourad Fakhfakh et al, Analog Circuit Design
Optimization through the Particle Swarm Optimization
Technique, Analog Integrated Circuits and Signal
Processing, 63:71-82, Aug. 2009.
[12] J. Kennedy, R. Eberhart, Particle Swarm Optimization,
IEEE International Conference on Neural Networks, 1995.
[13] J. Kennedy, The particle swarm: social adaptation of
knowledge, IEEE International Conference on Evolutionary
Computation, 1997.
[14] http://in.mouser.com/Passive-Components/ /N 5g73 [15]
S. Janson and M. Middendorf, A Hierarchical Particle
Swarm Optimizer and its Adaptive Variants, IEEE Trans.
Sys., Man, and Cyber. - Part B: vol.35, 1272- 1282, Dec.
2005.
Parameter Cap1 Cap 2 Cap 3 Cap 4 Cap 5
r () 0.10 0.10 0.10 4.83 0.10
l (nH) 11.16 0.01 0.01 10.14 0.01
c (nF) 0.001 3.12 17.87 0.001 3.12
x (mm) 0 0 0 0 200
y (mm) 150 150 0 150 150

You might also like