Professional Documents
Culture Documents
1158-3482-2-PB (STM JOURNALS Papers)
1158-3482-2-PB (STM JOURNALS Papers)
1158-3482-2-PB (STM JOURNALS Papers)
Abstract
In this model, the time domain waveform is evaluated for calculation of delay time, peak
time, settling time, damping ratio and natural frequency for a second order RLCG
interconnect network. It can also be used for multiple interconnect systems but for higher
order systems it is ignored due to accurate analysis. The model is applied to a single
resistance-inductance-capacitance-conductance model which can also be extended to multi-
interconnect systems to analyze the rise time and settling time in similar analysis. The
model evaluates the performance of a system which is expressed in terms of the transient
response for the unit impulse input because it is easy to generate and evaluate the delay
analytically. The transient response of a system to a unit impulse input depends upon the
initial conditions. In this paper, a new interconnect model is presented; the model is based
on the RLCG transmission line whose response is evaluated in time domain for a unit
impulse input. In this model the delay is calculated in SPICE and MATLAB. No
approximation is made to the transfer function of the interconnect. A closed form
expression for the propagation delay of a CMOS gate driving a distributed RLCG line is
introduced. On-chip inductance and conductance are shown to have a profound effect on
the high performance IC design methodologies. In this proposed model we have shown that
with the increase in the value of conductance by keeping constant the values of R, L, C we
evaluate that the SPICE delay reduces but if we compare it with the MATLAB proposed
delay model we see very accurately that the variation in the proposed delay is much larger
in comparison to the SPICE delay. Hence, for a high-speed circuit, one must increase the
value of G, so that the steady-state condition is reached as soon as possible. The simulation
results performed in Cadence SPICE environment justify the efficiency of the proposed
model.
Keywords: Time domain analysis, damping ratio, natural frequency, delay calculation,
RLCG interconnect, VLSI
Authors for Correspondence Email: abhi10091986@gmail.com, maheshwari_vikas1982@ya
hoo.com, rajibkarece@gmail.com
transmit a certain amount of data, thereby less than risetime/2, the line is electrically
increasing the performance of the system. short.
Global communication architecture based on
a global mesochronous, local synchronous A transmission line [5] can be described at
approach allows very high data rate per wire the circuit level using series inductance and
and therefore very high bandwidth in buses resistance combined with shunt capacitance
of limited width. This set of new challenges and conductance. An infinitesimal unit
is referred to as signal integrity in general. length of the transmission line looks like the
Among all these problems, capacitive circuit [19] in Figure 1. In Figure 1,
coupling induced cross talk is the issue that R = Series resistance per unit length
has been observed by an increasing number L = Series inductance per unit length
of backend vendors. Delay is determined by G = Shunt conductance per unit length
considering the time domain analysis on a C = Shunt capacitance per unit length.
transmission line after applying the input at
how much delay the output is being obtained.
Especially for an on-chip bus, delay is a
serious problem for VLSI design. In bus
structure, delay is most important because
long interconnect [14] wires often run
together and in parallel. Interconnect [15]
lines may be coupled to study the effects of Fig. 1 RLCG Parameters for a Segment of a
mutual inductive and capacitive coupling, Transmission Line.
such as delay. It is possible to use both a
distributed and a lumped model for these It is critical to model the transmission path
macro models. Pole [20] zero analysis is also when designing a high-performance, high-
being evaluated for finding the stability of speed serial interconnect system. The
the system. transmission path may include long
transmission lines, connectors, vias and
The rest of the paper is organized as follows: crosstalk from adjacent interconnect. Values
Section 2 discusses the basic theory, for R, L, C, and G are extracted from a given
transmission line model, crosstalk, glitch and geometry in 0.13 micron technology.
different modes of operation. Section 3
describes the difference model approach and B. Cross Talk
the proposed closed form formula for Crosstalk [13] is defined as the energy
bandwidth. Section 4 shows the simulation imparted to a transmission line due to signals
results. Finally, section 5 concludes the [23] in adjacent lines. The magnitude of the
paper. crosstalk induced is a function of risetime,
signal line geometry and net configuration
2. BASIC THEORY (type of terminations, etc.). In order to
overcome the problems faced at high
A. Transmission Line Models frequency of operation, shielding techniques
Defining the point at which an interconnect have been employed [11]. A common
should be treated as a transmission line and method of shielding is placing ground or
hence reflection analysis applied has no power lines at the sides of a victim signal
consensus of opinion. A rule of thumb is line to reduce noise and delay uncertainty.
when the delay [17] from one end to the The crosstalk between two coupled
other is greater than risetime/2, the line is interconnects is often neglected when a
considered electrically long. If the delay is shield is inserted, significantly
underestimating the coupling noise. The
ISSN: 2249 – 474X© STM Journals 2011. All Rights Reserved. 11
Journal of VLSI Design Tools & Technology
Volume 1, Issue 2-3, December 2011, Pages 10-19.
__________________________________________________________________________________________
C. Glitch
Crosstalk glitch (CTG) is a glitch signal
provoked by coupling effects among
interconnect lines which have unbalanced
drivers and loads [6]. The magnitude of the Fig. 2 An Example for Two Line
glitch depends on the ratio of coupling Transmission Line Model.
capacitance between line to ground
the frequency-domain models. The Simplifying the above two equations and
difference approximation procedure involves applying Laplace transformation, we get,
an approximation of the dynamic part of the V ( x)
( R sL) I ( x)
x (6)
system transfer function, given by Eq. (12), I ( x)
with the complex rational series or distorted (G sC)V ( x)
x (7)
part of the transient characteristic with the Differentiating Eqs. (6) and (7) with respect
real exponential series. This criterion results to x, and after simplifying we get,
in simple and efficient approximation 2V ( x)
2V ( x)
algorithms, and requires a minimal number x 2 (8)
of the original-function samples to be And
available, which is important if the line is 2 I ( x)
2 I ( x)
characterized with delay [9] and crosstalk x 2 (9)
aware bandwidth. where P is the propagation constant and is
defined as,
G. Modeling the Bandwidth Using Pd (R sL)(G sC)
Difference Model The general solution of Eq. (8) is given by,
We first consider the interconnect system V ( x) A1e x A2 e x (10)
consisting of single uniform line and ground Where A1 and A2 are the constants
as shown in Figure 5, and assume the length determined by the boundary conditions.
of the line is d. A differential length of a line From Eqs. (8) and (10), we get,
is assumed to possess distributed series
x
A1e x A2 e x ( R sL) I ( x)
inductance L and resistance R, as well as
shunt capacitance C and conductance G. After simplifying we get,
Kirchoff’s law is assumed to hold for the I ( x)
1
Z0
A1e x A2 e x (11)
small length (Figure 6).
where Z0 is the characteristic impedance.
Assuming that x = d, the termination voltage
and current are V(d) = V2 and I(d) = I2,
respectively, then we get,
V2 A1e d A2 e d (12)
1
I2 [ A1e d A2 e d ]
Z0 (13)
From Eqs. (12) and (13) we get,
A1 V2 I 2 Z 0 e d
1
2
A2
1
V2 I 2 Z 0 e d
2
Fig. 6 Equivalent Circuit of Each Uniform Substituting these values of A1 and A2 in Eq.
Section. (10)
V I 2 Z 0 ( d x ) V2 I 2 Z 0 ( x d )
V ( x) 2 e e
The electrical parameters of each section are 2 2 (14)
R∆X, L∆X, C∆X and G∆X, respectively, where Similarly we calculate for I(x) as,
R, L, C and G are per-unit length resistance, 1 V2 I 2 Z 0 ( d x ) V2 I 2 Z 0 ( x d )
I ( x) e e
Z 0 2 (15) 2
inductance, capacitance and conductance of
the line. Let at x = 0, V(x) = V1 and I(x) = I1 then from
Using Kirchoff’s voltage law, we can write, Eqs. (12) and (14), we can write,
V1 cosh(d )V2 Z 0 sinh(d ) I 2
v( x, t ) i( x, t ) Rx Lx
di( x, t )
v( x x, t ) (4) (16)
dt 1
I1 sinh(d )V2 cosh(d ) I 2
Using Kirchoff’s current law, we can write, Z0 (17)
dv( x x, t ) Since ABCD parameters are defined as
i( x, t ) Gx v( x x, t ) cx i( x x, t )
dt (5)
ISSN: 2249 – 474X© STM Journals 2011. All Rights Reserved. 14
Journal of VLSI Design Tools & Technology
Volume 1, Issue 2-3, December 2011, Pages 10-19.
__________________________________________________________________________________________
V1 A B V2 1 G( s ) 0
I C D I
1 2 (26)
So we can write ABCD matrix from Eqs. Now by using Eq. (24) the characteristic
(16) and (17) equation can be written as,
cosh(d ) Z 0 sinh(d )
V1 V2 LG RC RG
I 1 sinh(d ) cosh(d ) I
1 Z
s 2 s( )( 1) 0 (27)
0 2 LC LC
(18)
The output crosstalk voltage is given by, Now comparing the characteristic equation
V (s) (19) of second order system with unity feedback
V ( s) 1
2
cosh(d ) Eq. (27), we obtain the values of ω and
n
From Eq. (19), we can write the equation for ζ which are the natural frequency and
the transfer function of the system
damping ratios (Figure 7).
V ( s)
H ( s) 2
1 (20)
cosh(d ) RG 1
V1 ( s) (28)
After simplification of Eq. (20), we get the n LC
open loop transfer function, RC LG
V2 ( s ) 1 ζ LC (29)
G(s) (21) RG
V1 ( s ) R G (1 )
(s )(s ) LC
L C
plh phl
The above equation can be written as, d (30)
V2 ( s ) 2
4VDD VT ,n
1
G(s) (22) 2VT ,n
LG RC RG PHL
Cload
ln 1
{s s( ) K n VDD VT ,n VDD VT ,n
V1 ( s ) 2
} VDD
LC LC
For the stability analysis of the control (31)
system, we must find out the closed loop W
transfer function by considering the unity where K n n .COX n (32)
Ln
feed back with unit step input.
Consider the time response of second order 2 VT , p 4 VDD VT , p
ln 1
Cload
PHL
[22] system, in which we consider the
general form of closed loop transfer function
K p VDD VT , p VDD VT , p
VDD
which can be written as, (33)
V ω 2 (23) Wp
where K p p .COX
G( s) n
C ( s) 2 (34)
V 1 G( s) H ( s) s 2 2ζωns ω 2 L
1 n p
where n and ζ are the natural frequency Cload : Capacitive load applied to the output
and damping ratio. of the inverter
So the characteristic equation for this transfer COX : Gate-oxide capacitance
function can be written as,
VT : Threshold voltage for a transistor
1 G( s ) H ( s ) 0 (24)
VDD : Drain voltage applied to PMOS drain
By considering the open loop transfer
function with unity feed back, Eq. (22) can terminal.
be written as, p , n : Mobility of electrons and holes
V2 ( s ) 1 through transistor channel.
G(s) (25)
LG RC RG K n , K p : Transconductance of the NMOS
{s s( )( 1)}
V1 ( s ) 2
Table II RLCG Paramaeters for a Minimum-Sized Wire in a 0.18 µm Technology and the
Delay Model Parameters.
R L C G n ζ SPICE Proposed
(KΩ) (nH) (pF) (ms) rad/s delay delay
(1010 ) [td(ns)] [td(ns)]
1.2 2.7 2.4 1.5 1.66 26.81 1.46 27.25
1.2 2.7 2.4 2.25 2.03 26.83 3.99 24.35
1.2 2.7 2.4 3 2.35 26.84 2.58 21.00
1.2 2.7 2.4 3.725 2.62 26.86 0.68 18.85
In Table II we evaluate the performance of evaluate that the SPICE [16] delay reduces
the system by calculating the values of n
but if we compare it with the MATLAB
proposed delay model we see very
natural frequency ζ damping ratio at that accurately that the variation in the proposed
frequency. We calculate the MATLAB delay delay is much larger as compared to the
model for evaluation of time-domain SPICE delay.
analysis by using Eqs. (28) and (29). In this
proposed model we see that with the In Table III we have shown that with the
increase in the value of conductance by increase in the value of G the amplitude of
keeping constant the values of R, L, C we the impulse response increases and also the
time of that amplitude also increases but the peak overshoot, and the settling time is the
settling time reduces. In Table III, peak time time required for the response to reach and
is the time required for the response to reach stay within the specified range (2% to 5%)
the first peak of the time response or the first of its final value.
Table III. RLCG Parameters for a Minimum-Sized Wire in a 0.18 µm Technology and the
Time Domain Analysis Parameters.
x 10
8
Impulse Response
5
Peak amplitude : 4.79e+008
At time : 3.77e-011
4.5
Peak amplitude : 4.3e+008
At time : 4.2e-011
4
Peak amplitude : 3.72e+008
At time : 4.87e-011
3.5
Peak amplitude : 3.04e+008
At time : 5.95e-011
3
Amplitude
2.5
1.5
Fig. 10 A Spice Model for Delay
Settling time : 8.06e-009
Calculation in RLCG Network for the
1 Settling time : 8.98e-009
Values R(KΩ) = 1.2, L(nH) = 2.7,
Settling time : 1.04e-008
C(pF) = 2.4 and G(mS) = 1.5 in a 0.18 µm
0.5 Settling time : 1.27e-008
technology.
0
0 2 4 6 8 10 12 14
-9
Time (sec) x 10
Fig. 9 Impulse Response of a Linear Time
Invariant System.