Beruflich Dokumente
Kultur Dokumente
Madhav P. Desai
February 19, 2011
1 The Problem
Consider the following situation (Figure 1). Two nodes V and A are each being
driven by drivers and terminated by receivers. The load capacitances due to
receivers on the two nodes are CV and CA respectively. In addition, the two
nodes are connected by a coupling capacitance with value Cm .
We can identify two phenomena that result from the presence of the coupling
capacitance.
1. If the node V is being held to a fixed value (say 0V ) by its driver, and
if the node A switches, then a disturbance will be observed at V . This
disturbance may exceed the the noise margin at node V .
2. If the value at V is being switched, then the effective charge sourced from
(or sunk into) the driver of V will depend on whether A is switching or
not, and the direction of the switching at A. Thus, the delay of the driver
for V will vary depending on the activity at A.
2 Coupling noise at V
Consider the first phenomenon described above. The node A is switched from
low to high, and the node B is being maintained at 0V . The switching at node
A causes a disturbance at node V . If we assume that the disturbance at V is
small, then the switching of A will be at a rise-time approximately equal to
tA A
R = Rd (Cm + CA ) (1)
Assume that the node A switches as a linear ramp. An elementary analysis of
the situation shows that the waveform at V is given by
t
Cm RV − V
t A
VV (t) = (1 − e R ) (0 ≤ t ≤ t )
R (2)
tA
R
1
In other words, the disturbance at V is a pulse with height Vp 1 .
Examining the pulse-height Vp , we find that it can be written as
tA
tV − VR
Vp = KC R (1 − e t
R ) (4)
tA
R
1. If tVR /tA
R → ∞, then Vp → KC .
2. If tA V
R /tR → ∞, then Vp → 0.
3. If tA V
R = tR , then Vp → 0.63KC .
2
2.1 Reducing the number of sensitive signals
If the system is synchronous in nature, then the interference problem can be
contained to a great extent. For example, if the system has a single clock, and
if all state elements in the circuit are positive edge triggered, then it really
doesnt matter if there is coupling interference in most signals as long as they
have settled before they are sampled by a clock edge. In fact, only the clock
signal needs to be protected, and this can be done by shielding. Thus, choosing
an appropriate design paradigm (for example, synchronous positive edge trig-
gered) can simplify this problem considerably3 . One still has to pay attention to
interference effects in memories etc, but these are special and well-understood
cases.
3 Delay Variation
Note that the delay of the driver for V can vary in the range
Thus, the greater the coupling the greater the (relative) uncertainty in the delay
of the driver.
This delay variation in turn has two side-effects
1. The maximum clock-period in a synchronous system will need to be larger
than the worst-case maximum delay of a logic path. The delay variation
shown above will result in an overly conservative and hence inefficient
choice of clock-period.
3
V
Cc
References
[1] Vani Prasad, “Standardization of Interconnects: Towards an Interconnect
Library in VLSI Design”, Ph.D. Thesis, IIT Bombay, May 2006.