How Decoupling Capacitors May Cause Radiated EMI

Mark I. Montrose
Montrose Compliance Services, Inc.
[email protected]

Abstract–This paper analyzes effects that decoupling calculated from loop inductance (ESL) along with lumped
capacitor(s) may have on the development of radiated capacitance from discrete components. The need to minimize
emissions should improper implementation on a printed lead inductance is emphasized in the z-axis yet minimal
circuit board (PCB) occur. This applied EMC paper is based research has been presented on what happens when there is
on real-world experience and contains an easy solution to help excessive inductance on only one leg of a decoupling
engineers achieve compliance quickly and at low cost, without capacitor (i.e., power or ground-0V) [3, 4].
having to redesign the PCB. This paper presents what happens in a real-world PCB
Improper implementation of decoupling capacitor(s) should a poor layout topology be implemented. Not every
includes: routing traces on the outer layers or using boards
PCB layout designer, and in fact many digital design
with or without power and return planes; excessive
engineers, do not understand EMC and PCB layout
inductance in the decoupling loop area; lack of charge storage
to replenish the power and return planes; and self-resonant requirement and create a product using a routed trace between
frequency of the capacitor outside the harmonic spectrum of component and decoupling capacitor(s) because it is easy to
periodic signals. implement versus placing vias to a plane for every pin of the
We analyze the magnitude of radiated electromagnetic component and both legs of a capacitor.
interference (EMI) based on physical placement of decoupling Fast transient inrush surges causes a bounce condition to
capacitors from digital components and whether distance exist on both the voltage and 0V planes at switching
spacing from a switching element, connected by routed traces frequencies. In addition to plane bounce, if insufficient
on both the top and bottom of a PCB, makes a significant capacitance is present to minimize PDN noise, the magnitude
difference in the development and propagation of radiated of this bounce creates common-mode currents internal to the
emissions. The focus is on what occurs on a PCB, regardless of silicon of digital components. The impedance of the internal
whether the board is single-sided, double-sided or multilayer. bond wires to the package pins is extremely high upon which
common-mode current is also developed. Common-mode
Keywords-decoupling capacitors, trace inductance, power current now has several antenna structures to radiate from; the
distribution, common-mode radiation silicon or lead bond wires or routed trace(s) to a decoupling
capacitor. If there is insufficient energy charge in the PDN, a
I. INTRODUCTION potential plane bounce in power distribution may cause
common-mode current to be developed [3, 4, 5, 6].
Behavioral models chosen for simulation are usually Computational analysis may not include all parametric values,
(theoretically) perfect. Perfect models along with may be overly simplified, or not be well defined.
computational analysis may not represent an actual PCB
layout due to parasitics and other electromagnetic effects that II. TEST ENVIRONMENT AND SETUP
cannot be easily determined or anticipated. A voltage potential
difference due to inductance of the interconnect trace from a A simplified schematic and the PCB is shown in Figs. 1
decoupling capacitor to the power distribution network (PDN) and 2 to help in describing the test setup. A 74FCT244 is
can causes common-mode EMI to be developed and chosen based upon extreme operating parameters for which
propagated from digital components. this device is designed to be operated under. This component
Decoupling capacitors are required to minimize voltage has a large output voltage drive level (+5V CMOS), fast edge
potential differences and plane bounce along with RF transition rate and up to 960 mA data book inrush current
switching energy injected into the PDN from digital consumption under maximum capacitive load. A clock signal
components. Once switching energy is injected into a PDN, at different frequencies is applied to all eight input gates
this energy will be distributed throughout the power/return simultaneously, stressing the component to maximum
network within a certain radius of operation and can cause operating/thermal parameters. Each output driver has a
functional disruption of circuits should the magnitude of the maximum resistive/capacitive load (47pF/47). When all
plane bounce exceed operational margins. The magnitude of eight drivers switch simultaneously, the inrush surge current
radiated energy, related to the physical placement of a was 380 mA. The resistor/capacitor pair was connected
decoupling capacitor with routed traces is investigated herein. directly to each output, using a single via to minimize routed
Prior research on the effectiveness of decoupling trace inductance. This routed distance was 1.6 mm (0.062
capacitors is extensive and covers namely the effects of lead inches or 62 mils).
inductance and RF switching noise present within a PDN, but Provisions for six decoupling capacitors with different
not radiated emissions from a routed trace between component trace lengths were provided to analyze the magnitude of
and capacitor. Examples include [1, 2]. It is a well-known fact common-mode current developed internal to the component’s
that a PDN must have low impedance at switching frequencies package and to determine whether the decoupling capacitor
to minimize plane bounce. Plane impedance is easily
location would increase radiated EMI. One configuration had III FREQUENCY DOMAIN ANALYSIS
a capacitor (C16) located immediately adjacent to the input
power pin, or best possible location. Five capacitors (C11- The reason 3.9nF capacitors were originally chosen is due
C15) were located at a distance of 6.4, 7.6, 8.9, 10.2 and 11.4 to how a typical junior design engineer may [incorrectly]
cm (2.5, 3.0, 3.5, 4.0 and 4.5 inches) away. A non-typical select a decoupling value without taking into consideration
capacitor value (3.9 nF) was chosen to analyze what may loop inductance. The self-resonant frequency of a capacitor at
occur if there is insufficient decoupling capacitance instead of 80 MHz, with typical lead inductance (ESL) of 1nH is 3.9nH,
a theoretically perfect 100 nF device. Eq. (1). This capacitor selection value proved to be interesting,
Power trace as discussed later. If we used the correct value of ESL, the
Battery input 74FCT244
self-resonant frequency along with other details provided in
Signal
input Table 1 now makes sense. Trace inductance of a microstrip
transmission line on this PCB stack is 7-nH/cm (18-nH/inch).
To other
capacitors 1 (1)
located at f 
various 2 ESL  C
distances
away from
component. Radiated emission, E (V/m) from the loop area between
capacitor and component can be easily determined by Eq. (2)
Ground trace [7]. Current draw is 380 mA under maximum load conditions.
The variables in (2) involve f = frequency in MHz, A=loop
area (m), Idm = current (mA), and r = distance between PCB
Two topologies investigated. and antenna (meters). Three meters is used to represent FCC
Class B limits.
Fig. 1. Simplified schematic of the test PCB
1
E  263  10 16 ( f  A  Idm)  (2)
r
Equation (3) calculates the wavelength of a propagating
signal and is related to total loop perimeter from capacitor
placement, with f = frequency (MHz) and  = wavelength
(meters). This is the frequency in which the loop area becomes
an efficient antenna. Consequently, any loop size from
decoupling capacitor placement being very small, cannot
become an efficient radiating differential-mode or loop
antenna. Test data and plots confirmed this.

300
f  (3)

The physical location of a decoupling capacitor, if routed
with a trace to a component, allows a loop to exist. This loop
does have an effect on radiated emissions, but not based on the
Fig. 2. PCB layout actual physical size of the loop.
Amplitude levels from all decoupling locations were
Three different decoupling capacitance values were nearly identical when measured in a TEM cell, indicating size
investigated separately and then in groups of five: 1nF, 3.9nF of the loop did not have significant effect on radiated EMI.
and 100 nF. Zero ohm resistors were used to extend the length Radiated emission from the PCB with different
of the decoupling loop. These resistors have a minimal amount decoupling capacitor values is provided in Fig. 3. Regardless
of equivalent series resistance (ESR) and equivalent series of decoupling loop size, radiated emissions were exactly
inductance (ESL), and thus did not affect measurement results. identical which indicates that the size of a decoupling loop
Ferrite beads prevented switching noise corruption to/from the makes no difference in EMI, since EMI is being radiated with
DC power source (battery). significant magnitude from somewhere else and not from the
An 80 MHz clock signal (maximum operating frequency loop or its physical size.
of the 74FCT244) was provided using a SMA connector. The One reason why a decoupling loop may cause radiated
reason to not provide on-board clock generation was to allow EMI is explained by [6], which had a piece of wire attached to
measurement of actual emissions from the 74FCT244 by a plane driven by common-mode currents present within the
itself, along with the effects from various decoupling loop PDN. Ref. [6] did not investigate traces routed microstrip, the
areas; not from a radiating frequency generation source. Due focus of this paper. When a transmission line (trace) is
to an extensive amount of data (40 plots), typical results can energized with common-mode current created as a result of
only be presented. The key item to observe is both cause and insufficient decoupling or interplane capacitance, common-
effects based on a many test plots and data. mode currents developed internal to the component now have
an antenna to propagate EMI from.
 Table 1. Parametric Details of the Test PCB
Capacitor Loop perimeter due to Total loop Self-resonant frequency Calculated emission Calculated radiated
ID capacitor placement inductance 1 of the loop path3 levels @ 10 MHz4 frequency (loop area)5
C11 2.5 inches (6.4 cm) 46 nH 11.5 MHz 108.8 V/m 4.8 GHz
C12 3.0 inches (7.6 cm) 55 nH 10.8 MHz 153.4 V/m 3.9 GHz
C13 3.5 inches (8.9 cm) 64 nH 10.0 MHz 210.3 V/m 3.4 GHz
C14 4.0 inches (10.2 cm) 73 nH 9.4 MHz 276.3 V/m 2.9 GHz
C15 4.5 inches (11.4 cm) 82 nH 8.9 MHz 345.1 V/m 2.6 GHz
C16 0.01 inches (0.25 mm) 3.18 nH 2 45.2 MHz 0.17 V/m 1,200 GHz
1. Includes package lead-length inductance (1-nH) and trace inductance from the total loop area (18 nH/inch).
2. Inductance value contains lead-length inductance (1-nH) and one via (2-nH).
3. Value of capacitor used for calculating self-resonant frequency, Eq. (1): 3900 pf (3.9 nF) and total loop inductance, L.
4. Distance used is 3 meters, calculated from Eq. (2).
5. Loop area determined from loop perimeter for use with Eq. (3).

Common-mode current
Source
Decoupling capacitor
Decoupling (dielectric emulates free space)
loop

Ground Power
Power Ground
Dipole structure Source Ground plane

Fig. 4. Antenna structure due to decoupling placement

No decoupling capacitor – 10 MHz
The loop area on PCBs between a capacitor and component
is physically small, which means they become an efficient
radiating loop antenna in the GHz range. Decoupling capacitors
do not provide decoupling in this frequency range. Decoupling
capacitors instead creates a dipole antenna structure but only if
there is a routed trace and not via directly to a PDN. We must
concern ourselves with common-mode current developed from
driving a dipole antenna.
The self-resonant frequency of a 3.9 nF, surface mount
capacitor (0805 package size) with lead length inductance of 1-nH
Worst case decoupling, 3.9 nF–11.4cm (4.5"), 10 MHz is 80 MHz, as long as the capacitor is never installed on the PCB.
When installed on the PCB, lead inductance reduces the self-
resonant frequency from 80 MHz to approximately 10 MHz.
Investigation occurred at both 80 MHz and 10 MHz.
Radiated emissions were measured with different
capacitance value to determine differences in radiated emissions
from the size of the loop. Fig. 4 shows replacing the 3.9 nF
capacitor (chosen because it was supposed to be self-resonant at
80 MHz) with a 100 nF capacitor. The 100 nF capacitor is self-
resonant at 16 MHz, using actual trace inductance. The 3.9 nF
capacitor, located in a best-case physical position (C16), is
resonant at 45 MHz. The 3.9 nF capacitor is self-resonant at
Typical decoupling location, 3.9 nF–6.4cm (2.5") 10 MHz
approximately 10 MHz at locations C11-C15.
If physical placement has no effect on emissions, then why
Figure 3. Radiated emissions from sample loop areas.
was less radiated EMI observed with the 100 nF capacitor? This
was the result of having a greater amount of stored electrical
A dipole antenna is created by the existence of the
charge provided by the larger value capacitor, not loop area
decoupling loop, with the driven element at voltage potential and
dimension. The 3.9 nF capacitor, with limited energy storage
the ground element at 0V. When any capacitive element is
operating at 45 MHz, could not provide sufficient energy to the
installed between the two elements of a dipole antenna, common-
component operating at both 10 MHz and 80 MHz. Figure 5
mode current propagates through the capacitor which closes the
illustrates effects of five identical value decoupling capacitors in
RF loop for AC currents. The capacitor's dielectric emulates free
parallel to increasing charge storage. The net effect is that the
space conditions, allowing common-mode current developed
total amount of energy to the 74FCT244 is five times greater. The
internal to a component to drive this virtual dipole antenna if
greater the charge energy at the operating frequency of the device,
plane bounce occurs. This concept is illustrated in Fig. 4 [3, 4].
a reduction in radiated emission is observed. Reduced EMI occurs
because the component produces less common-mode current since
the PDN is now providing sufficient charge to keep the planes
from bouncing due to lack of energy storage.

Five capacitor – 1 nF (5 nF total) – 10 MHz

Minimum loop dimension possible, 3.9 nF – 10 MHz

Five capacitor – 3.9 nF (20 nF total) – 10 MHz

Minimum loop dimension possible, 0.1 F - 10 MHz

Fig. 4. Effect of different capacitor values, same location

III. CONCLUSION

The dimension of a decoupling loop, due to capacitor

placement, does play a role in amplitude development and Five capacitor – 100 nF (500 nF total) – 10 MHz
propagation of RF emissions. The perimeter of the loop emulates
a dipole antenna, radiating common-mode energy, with the Fig. 5. Magnitude of emissions - multiple capacitors
magnitude of EMI identical regardless of loop size. The capacitor
acts as the driven portion of the antenna structure due to an 2. T. P. Van Doren, J. Drewniak, T. H. Hubing. "Printed Circuit Board
imbalance in the differential-mode power distribution network if Response to the Addition of Decoupling Capacitors," Tech. Rep.
both power and return traces are not exactly identical in length. A #TR92-4-007, University of Missouri, Rolla EMC Lab., 1992.
very small loop does not generate any greater magnitude of (September 30).
radiated emissions than a capacitor located further away with a 3. Montrose, M.I. Analysis on Loop Area Trace Radiated Emissions
from Decoupling Capacitor Placement on Printed Circuit Boards.
larger loop size although spectral shifts are observed. This
Proceedings of the IEEE International Symposium on
situation is valid as long as any trace exists on either the top or Electromagnetic Compatibility, 1999. 423-428.
bottom of the PCB, but does not apply if the capacitor is 4. Montrose, M.I. 2000. Printed Circuit Board Design Techniques for
connected directly to planes by vias to minimize lead inductance. EMC Compliance. Wiley/IEEE Press.
If sufficient amount of charge storage is provided, the silicon 5. Paul, C.R. "A comparison of the contributions of common-mode and
internal to a component package will not bounce planes to any differential-mode currents in radiated emissions." IEEE Transactions
degree. When planes bounce occurs common-mode current is on Electromagnetic Compatibility 31, 2:189-193.
developed which now sees a dipole structure to drive the trace. To 6. Drewniak, J. L., T. H. Hubing, T. P. Van Doren. 1994.
"Investigation of Fundamental Mechanisms of Common-Mode
prevent this situation from occurring, never use a routed trace
Radiation from Printed Circuit Boards with Attached Cables."
between component and decoupling capacitor. Proceedings of the IEEE International Symposium on
If EMI occurs that cannot be easily identified from where the Electromagnetic Compatibility, 1989. 110-115.
energy is coming from, and if there is a decoupling capacitor with 7. Ott, H. 1988. Noise Reduction Techniques in Electronic Systems,
a routed trace, remove this capacitor. There is probably sufficient 2nd ed. New York: John Wiley & Sons, Inc.
amount of decoupling already present and removing one capacitor 8. Tang, G. "Surface Mount Capacitor Loop Inductance Calculation
may not cause functional disruption. and Minimization." Proceedings of the IEEE International
Symposium on Electromagnetic Compatibility, 1998. 505-510.
9. Yuan, F. "Analysis of Power/Ground Noises and Decoupling
REFERENCES Capacitors in Printed Circuit Board Systems." Proceedings of the
IEEE International Symposium on Electromagnetic Compatibility,
1. Drewniak, J. L., T. H. Hubing, T. P. Van Doren, D. M. Hockanson. 1997. 425-430.
"Power Bus Decoupling on Multilayer Printed Circuit Boards."
IEEE Transactions on Electromagnetic Compatibility 37(2), 1995.
155-166.