0% found this document useful (0 votes)
70 views88 pages

Practical Power Solutions

Uploaded by

nomiman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
70 views88 pages

Practical Power Solutions

Uploaded by

nomiman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 88

Hardware Design Techniques

Practical Power Solutions


1. Point-of-Load Power
2. System Power Management and Portable Power
3. Power for Mixed Analog/Digital Systems
4. Hardware Design Techniques

Copyright © 2009 By Analog Devices, Inc.


All rights reserved. This book, or parts thereof, must not be
reproduced in any form without permission of the copyright owner.

SECTION 4
HARDWARE DESIGN TECHNIQUES

Passive Components.......................................................................................................4.1
Active Components.......................................................................................................4.39
Power Supply Layout and Grounding.........................................................................4.52
Shielding........................................................................................................................4.73
Thermal Design............................................................................................................4.76
Technical References....................................................................................................4.84

4.0
Hardware Design Techniques

Passive Components

Passive components play a significant role in the operation of switch mode power supplies (SMPS).
Inductors are the primary energy storage device in most SMPS. Capacitors are used for filtering,
decoupling, energy storage, and affect the design of the compensation network since the SMPS is a
closed-loop feedback system. Resistor dividers are often used to set the output voltage, and low value
resistors are often used in current sense applications.

4.1
Hardware Design Techniques

Capacitors

‹ A capacitor is an energy storage element constructed of 2 conductors


separated by an insulating material
A 1 2
C = ε ⋅ ε0 ⋅ ESTORED = ⋅C⋅ V d
d 2
A
‹ Where
z ε0 is the dielectric constant of free space
z ε is the relative dielectric constant of insulator
z ε is sometimes called the "k-factor" or simply "k"
z A is area of conductive plates
z d is distance between plates
z V is the voltage potential across plates
‹ Larger capacitors have larger capacitance, and therefore better energy
storage.
‹ Different capacitors use different dielectric material. This changes
capacitance and their characteristics (ESR, Current Rating, DC Bias, etc)
in power designs

This figure shows the basics of capacitors.


Capacitors come in a wide variety of sizes, both in capacitance value and physical size. Choosing the
right capacitor for a specific application can be crucial to the proper operation of the circuit. Choosing
the right capacitor means not only choosing the correct value but also the right dielectric material as
well. As we will see in the following pages, real world capacitors are far from ideal, especially over a
wide frequency range.
The capacitance can be calculated from the information in this figure. It is dependant on the dimensions
of parallel conductors, the separation of the conductors, and the dielectric constant of the insulator. The
capacitive impedance can be calculated from:

ZC = 1/jωC
where:
j = √(–1)
and
ω = radian frequency = 2π · frequency in Hertz.

The energy stored in the capacitor which is charged to a voltage, V, is given by the formula:

E = 0.5CV2

4.2
Hardware Design Techniques

A Real Capacitor Equivalent Circuit


Includes Parasitic Elements

Ignore for power designs


RP

RS L
C

ESR ESL

RDA CDA

This is a workable model of a non-ideal capacitor. The nominal capacitance, C, is shunted by a


resistance, RP, which represents insulation resistance or leakage. A second resistance, RS, (equivalent
series resistance, or ESR), appears in series with the capacitor and represents the resistance of the
capacitor leads and plates.
Note that elements in the capacitor equivalent circuit aren't that easy to separate. The model is for
convenience in explanation. Inductance, L (equivalent series inductance, or ESL), models the inductance
of the leads and plates. Finally, resistance RDA and capacitance CDA together form a simplified model of
a phenomenon known as dielectric absorption, or DA. This can degrade fast and slow circuit dynamic
performance. In a real capacitor, RDA and CDA may actually consist of multiple parallel sets.
In most cases the values of RP, RDA, and CDA are more important in high frequency (RF) applications.
We will neglect them in most power supply applications and focus on C, ESR, and ESL.

4.3
Hardware Design Techniques

"Bathtub" Impedance of 100µF Capacitors


100

10 C
ALUMINUM
ELECTROLYTIC
ESL ESL = 16nH
1
Z Ω ESR
ESL = 1.6nH
100m
CERAMIC
MLCC
10m

1m
100 1k 10k 100k 1M 10M
FREQUENCY (Hz) ESR → 0
1 1
Self-Resonant Frequency =
2π ESL×C

Theory tells us that the impedance of a capacitor will decrease monotonically as frequency is increased.
Practice tells us that at some frequency the ESR will dominate, and the impedance plot will flatten out.
In fact, as we continue up in frequency, the impedance will start to rise. This is where the ESL starts to
dominate. Where these "knee" points occur will vary with capacitor construction, dielectric, and value.
This is why we often see larger value capacitors paralleled with smaller values. The smaller value
capacitor will typically have lower ESL and continue to “look” like a capacitor higher in frequency. This
improves the overall performance of the parallel combination over a wider frequency range.
The self-resonant frequency of the capacitor is the frequency at which the reactance of the capacitor,
1/ωC, is equal to the reactance of the ESL, ω×ESL.
This figure compares the impedance of a 100 µF aluminum electrolytic capacitor with a 100 µF multi-
layer ceramic capacitor (MLCC). Note that the ceramic capacitor has a much lower ESL and ESR than
the aluminum electrolytic. Therefore the "dip" at the ceramic capacitor's self resonant frequency is much
more pronounced than the dip in the aluminum electrolytic.
An ideal capacitor with ESR = 0 has an infinite "dip" at the self-resonant frequency as shown in the
figure.

4.4
Hardware Design Techniques

Response of a Capacitor to a Current Step


i IPEAK = 1A

INPUT
v di 1A
CURRENT =
dt 100ns
STEP
ESR = 0.2 Ω
0 Equivalent f = 3.5MHz

ESL = 20nH ESL "Spike"


di
VPEAK = ESL × dt + ESR × IPEAK
= 400mV
C = 100µF OUTPUT
XC = 0.0005Ω VOLTAGE
ESR × IPEAK = 200mV
@ 3.5MHz
ESR "Step"
0

The complex impedance of a “real” capacitor will also cause non-ideal response to a current step. The
assumption here is that we are stepping from a steady state current of 0 A to a steady state of 1 A. We
also assume that the rise time of the current step is about 100 ns, which corresponds to a frequency of
about 3.5 MHz. A “perfect” 100 µF capacitor has an impedance of 0.5 mΩ at 3.5 MHz.
We would expect to see the capacitor voltage start at zero (the voltage across an ideal capacitor cannot
change instantaneously) and then slew at a rate equal to IPEAK/C (which in this case works out to be
10,000 V/s, or 10 mV/µs). The ideal capacitor should have only a small voltage change during the
100xns rise time of the current step. (Approximately 1 A × 0.0005 Ω = 500 µV).
The actual response of the non-ideal capacitor (ESR = 0.2 Ω, ESL = 20 µH) is shown in the figure.
The fast-rising edge of the current waveform shown results in an initial voltage peak across the
capacitor, which is proportional to the ESL. The current through the inductor cannot change
instantaneously. After the initial inductive transient, the voltage settles down to a longer duration level
which is proportional to the ESR of the capacitor. Thus the ESL determines how effective a filter the
capacitor is for the fastest components of the current signal, and the ESR is important for longer time
frame components.
Note that an overall time frame of a few microseconds (or even less) is relevant here. As things turn out,
this means switching frequencies in the 100 kHz to 1 MHz range. Unfortunately this happens to be the
region where many switching supplies operate, and the ESR and ESL effects become critical.

4.5
Hardware Design Techniques

Capacitor Data Sheet : What Do Specs Mean ?

‹ Rated Voltage: Maximum voltage for which the capacitor is guaranteed to function. Often there is no
guarantee of actual capacitance at this voltage.
‹ Rated Temperature: Maximum operating temperature. Temperature rise of capacitor based on ripple
current should be included.
‹ Rated Capacitance: Usually specified to ±20% or ±30%. Note this is the capacitance at no dc bias and at
low frequency. Always look at data sheet plots to check predicted capacitance at your design conditions
(especially dc bias).
‹ Dissipation Factor, DF: This is a measure of the power loss of the capacitor expressed as a %. DF =
2πfRC×100%, where R = ESR, f = frequency (Hz)
‹ Leakage Current, LC: The amount of leakage current after a specified amount of continuous operation (5
to 10 minutes). Leakage current will cause the capacitor to discharge itself with no load attached.
‹ Equivalent Series Resistance, ESR: Includes conductor resistance, dielectric loss, and leakage. This is a
very important specification, as it will affect load transient, ripple voltage at input and output, and
maximum current capability.
‹ Maximum Ripple Current: The maximum allowed RMS current in a dc-to-dc converter application. The
frequency is usually specified at 100kHz.

This figure defines the terms that you may find in a specification page for a capacitor.
Different value ranges of capacitors may have different specifications—the one shown is for a large
value electrolytic. Small value capacitors may emphasize different specifications. Also you may find that
capacitors targeted at different frequency applications may have ESR, etc., specified at frequencies other
than 100 kHz.
We will now examine the functions of capacitors in switch mode power supplies.

4.6
Hardware Design Techniques

Capacitors for Power Designs: Energy Storage


BOOST CONVERTER (ON CYCLE)
VIN D1 VOUT
ΔVOUT
L1 tON
D1 (OFF)

CIN ΔVOUT
COUT ILOAD = C ⋅
t on
ILOAD
FROM SW 1 (ON) I
PWM
ΔVOUT = t on ⋅ LOAD
COUT
CONTROL

‹ Energy stored in a capacitor increases as voltage is applied across it.


‹ It can provide energy quickly as required—acts as an energy reservoir.
‹ If load changes, capacitor will supply energy until loop can react. Larger
capacitor will give better regulation.
‹ Larger output capacitor = less voltage ripple, neglecting ESR effects

This figure shows how the COUT output capacitor acts as an energy storage device in a boost converter.
During the time SW1 is on and D1 is off, COUT must supply all of the current to the load. The voltage
will "droop" at a rate equal to ILOAD/COUT until the next charge cycle (SW1 off, D1 on) replenishes this
charge.
The choice of the value for COUT is a tradeoff—trying to balance capacitor (physical size), cost, and
ripple current.
Since the output ripple of a boost converter current is high, COUT needs to be rated to handle the required
ripple current. This generally means that the ESR will be low.
A good starting point for COUT size is to use the next standard value up from the value calculated from:

I
COUT = t ON LOAD
ΔVOUT

4.7
Hardware Design Techniques

Capacitors for Power Designs: Filtering


vD1 BUCK CONVERTER
iSW
VIN 0
vD1
L1 VOUT
0

iSW
iL1 ⎛ 1 ⎞
ΔV ≅ ΔI × ⎜⎜ + ESR COUT ⎟⎟
PWM ⎝ 8 × f × COUT ⎠
CIN COUT
ESR
D1
ESR
ILOAD

ESL
ESL

‹ The LC filter at the output of the buck converter filters the square wave at the
switch node, vD1.
‹ The amount of voltage ripple seen at VOUT is inversely proportional to the
capacitor value and proportional to ESR.
‹ We can make C very large, so the ultimate limit is the ESR.
‹ The input filter capacitor will reduce the noise injected onto the VIN line which can
interfere with other parts of the system.

The second function of capacitors in switch mode power supplies is filtering. This shows how the COUT
capacitor in a buck converter is used to reduce the output ripple voltage.
In this instance the inductor will smooth out the ripple, since the current in an inductor can’t change
instantaneously. While that may seem like a good thing, it should be remembered that low ripple and
good transient response are mutually opposed to each other. Also the regulator is a closed loop negative
feedback system, so the laws governing stability must be obeyed. The Bode plot of the system needs to
be examined for adequate phase margin.
The input current of a buck converter is discontinuous, therefore the input filter capacitors must reduce
the noise injected onto the VIN line which can interfere with other parts of the system. The peak source
impedance should be about three times lower than VIN/IIN.
This is generally achieved by keeping the input capacitor ESL small and CIN large. Sometimes the CIN
ESR has to be decreased to lower the input impedance, or two capacitors in parallel are used for CIN—
one with small capacitance and low ESR/ESL and one with large capacitance and higher ESR/ESL.

4.8
Hardware Design Techniques

Boost Converter with Output Pi Filter


OUTPUT PI FILTER

VIN L1 D1 L VOUT

CIN
COUT C

ILOAD

The boost converter has a discontinuous pulsed output current which is more difficult to filter than the
buck converter output. Rather than try and reduce the ripple with a single output capacitor COUT, it is
often more efficient to use an output pi filter as shown in shaded area of the figure. The output noise
filter inductor, L, reduces output ripple voltage by attenuating the ac current passing through the output
noise filter capacitor, C. The output inductor, L, is subjected to low ac voltage and very low ac current.
It is generally safe to use an inexpensive drum core or "open field" type inductor in this application. The
output noise filter capacitor, C, should have low ESL and ESR but does not handle much ac ripple
current.
Be careful in applying pi output filters as they generally degrade output load transient response. If the
filter is placed inside the feedback loop, compensation will be tricky. The pi filter can be equally
effective when used on the output of a buck-boost converter.

4.9
Hardware Design Techniques

Capacitor Functions in Buck Converter


SYNCHRONOUS BUCK CONVERTER,
DISCONTINUOUS INPUT CURRENT
CONTINUOUS CONDUCTION MODE
VIN DC + RIPPLE
CONTINUOUS OUTPUT CURRENT
0
VIN
DC + RIPPLE
iSW2
CIN 0
SW2 iL1 VOUT
RIPPLE ONLY
L1 COUT
ESR
DC + RIPPLE
iSW1 RIPPLE ONLY ESR
ESL iCIN SW1 iCOUT
ESL ILOAD
0
DC + RIPPLE

‹ CIN must handle high discontinuous ripple current. ESR, ESL should be low.
‹ COUT has continuous ripple current and must minimize output ripple voltage. ESR
should be low.
‹ COUT affects loop stability and load step transient response.

The input ripple current waveform of a buck converter is discontinuous with high di/dt and peak-to-peak
amplitude. The input capacitor supplies pulse currents to the upper MOSFET. The input capacitor must
therefore have high ripple current handling capability and low inductance. It is usually unwise to allow
this input ripple current to be handled by other, uncontrolled bypass capacitors in the system. An
inductor is sometimes placed in series with the input supply line capacitor to provide further filtering.
Output ripple current in a buck converter is continuous and usually low, so capacitor ripple current
ratings are not usually an issue. However, buck regulators are commonly used to power loads such as
processor cores and FPGAs which can impose severe dynamic load regulation requirements. When
processor core output voltages approach 1 volt, the allowed error band is proportionately small (50 mV
for 5% error), so a controlled low impedance over a broad frequency range is required.
The output capacitor is an integral part of closed loop stability of the buck converter. It smoothes the
ripple current coming out of the converter, reducing the ripple voltage. The output capacitor supplies
current during transient conditions until the feedback loop acts to increase the inductor current.
A good design tool, such as ADIsimPower, calculates the ripple current in the capacitors and inductors
and makes the appropriate parts selection.

4.10
Hardware Design Techniques

Capacitor Functions in a Boost Converter


DISCONTINUOUS OUTPUT CURRENT
CONTINUOUS INPUT CURRENT
VOUT – VIN
DC + RIPPLE DC + RIPPLE
L1
VIN 0
VOUT
iL1 D1
iD1

CIN iSW1
COUT
iCIN
ESR ESR
ILOAD
RIPPLE ONLY
ESL SW 1 RIPPLE ONLY
ESL

‹ CIN filters continuous ripple current and stabilizes feedback loop.


‹ COUT filters discontinuous output ripple current and must minimize
output ripple voltage. ESR should be low.
‹ Pi-Filter on output often less costly than optimizing COUT.
‹ COUT affects loop stability and load step transient response.

As previously discussed, the boost converter has continuous input current and discontinuous output
current. The input capacitor smoothes the ripple current going into the converter, reducing the ripple
voltage on the input. The input capacitor also provides low impedance on the input to ensure stability.
The output capacitor is an integral part of closed loop stability of the boost converter. It must supply the
load current when the MOSFET is on and the diode is off.
Output capacitor also supplies current during load transient conditions until the feedback loop responds
and increases the inductor current.
Depending upon the capacitor technology, an output filter that is designed to handle the high ripple
current will tend to produce high output ripple voltage. An output filter that is designed to produce the
desired ripple voltage is likely to be significantly larger and more expensive. So boost converters often
use a pi filter in the output to more efficiently filter the output ripple voltage.

4.11
Hardware Design Techniques

Capacitor Functions in Buck-Boost Regulators


DISCONTINUOUS INPUT CURRENT

VIN DC + RIPPLE
0
VIN DC + RIPPLE
iSW1
–VOUT
CIN SW1 iL1
RIPPLE ONLY L1 COUT
ESR RIPPLE ONLY
DC + RIPPLE
ESR
iCIN ILOAD
ESL iCOUT
ESL
iD1
–VOUT
D1
DISCONTINUOUS OUTPUT CURRENT

‹ Large discontinuous input and output ripple current with high di/dt.
‹ Often requires a pi filter on the output to control ripple.

In terms of input/output ripple current, the buck boost converter offers the worst of both worlds—large
discontinuous ripple current with high di/dt on both the input and output, both must be handled by their
respective filter capacitors.
Like the boost converter, the buck-boost converter is often used with a pi filter on the output, instead of
the simple “L” section shown here.

4.12
Hardware Design Techniques

Parallel Capacitors

C C 2C

ESR ESR ESR


2

ESL ESL ESL


2

‹ Twice the capacitance, half the ESR, half the ESL


‹ Sometimes more cost effective than trying to do it with a single capacitor
‹ Sometimes the only way to get required C and ESR to meet ripple
requirements
‹ May increase real estate, but depends highly on type of capacitor
‹ Bulk capacitors may be composed of several parallel capacitors,
sometimes of different values

Output capacitors are often used in parallel combinations of multiple individual capacitors. Parallel
capacitors have a total capacitance equal to the sum of all their individual values.
An added benefit is that the effective ESR and ESL is reduced since the impedance of paralleled
resistors goes according to the formula:

1
=
1
+
1
+ …+ 1
ZT Z1 Z2 ZX

In some cases parallel capacitors provide a more cost effective solution than trying to use a single
capacitor. Sometimes parallel capacitors are the only way to get the required C and ESR to meet the
ripple requirements.
It is a common practice to parallel different types and values of capacitors to take advantage of the low
ESR and ESL of ceramic capacitors, while having the benefit of large capacitance values of aluminum
or "bulk" capacitors. However, when you parallel different types of capacitors, you can't simply add the
C value and parallel the ESR and ESL values. The resulting impedance must be calculated based on the
complex impedance model.

4.13
Hardware Design Techniques

Popular Capacitor Types


TECHNOLOGY ADVANTAGES DISADVANTAGES APPLICATIONS

Aluminum •High CV •Temperature related •Consumer products


Electrolytic product/cost wearout •Large bulk storage
(Switching Type) •Large energy storage •High ESR/size
•Best for 100V to 400V •High ESR @ low temp
Solid Tantalum •High CV product/size •Fire hazard with reverse •Popular in military
•Stable @ cold temp voltage •Concern for tantalum
•No wearout •Expensive raw material supply
•Only rated up to 50V
Aluminum-Polymer, •Low ESR •Rapid degradation •Newest technology
Special-Polymer, •Z stable over temp above 105°C •CPU core regulators
Poscap, OsCon •Relatively small case •Relatively high cost

Ceramic •Lowest ESR, ESL •CV product limited •Excellent for HF


•High ripple current •Microphonics decoupling
•X7R good over wide •C decreases with •Good to 1GHz
temp increasing voltage
Film (Polyester, •Hi Q in large sizes •CV product limited •High voltage, current
Teflon, •No wearout •Not popular in SMT •AC
polypropylene, •High voltage •High cost •Audio
polystyrene, etc.

This is a survey of some of the more popular capacitor types. Note that not all values are available in all
types, and the physical size can vary widely for the same value of capacitance.

4.14
Hardware Design Techniques

Aluminum Electrolytic Capacitors

‹ Aluminum electrolytic capacitors provide large capacitance (100µF to >


1mF) in relatively small size at very low cost
‹ Represent the best µF/cost of all options
‹ Achieve this with high dielectric constant (ε) due to oxide film, and thin
layers, (d is small)
‹ High ESR: can be several ohms
‹ Aging due to dry-out must be taken into account in long lifetime designs
‹ Surface mount available, but some reliability issues. Other technologies
better for lower capacitance values
‹ Poor high frequency performance
‹ Use "switching" types, not "general purpose" in switcher designs

Switching type aluminum electrolytic capacitor are designed and constructed for use in SMPS. They
have lower ESR and ESL than general purpose aluminum capacitors, generally meaning higher ripple
current. "General purpose" aluminum electrolytic capacitors should not be used in switcher designs.
Aluminum electrolytics are available in large values and in relatively small size and low cost. This is
accomplished by using an oxide film with high dielectric constant and thin layers.
The chief disadvantage of aluminum electrolytics is the high ESR, which can be several ohms.
Because of their poor high frequency performance, they are often paralleled with other types of
capacitors to achieve the desired filtering characteristic.

4.15
Hardware Design Techniques

Solid Tantalum Capacitors

‹ A type of solid electrolytic capacitor, using tantalum powder as dielectric


‹ Size efficient for 10’s of µF range
‹ Capacitance is temperature stable
‹ ESR is in the range of 100’s of mΩ, lower than aluminum electrolytics
‹ Must be designed for 2 × voltage rating
‹ Safety is a concern (e.g. reverse polarity, overvoltage rating): as failure is
seen as burn out of capacitor with flame
‹ Popular in high-rel military applications
‹ However, becoming less popular, as ceramic capacitors can achieve
same capacitance with lower ESR, and ESL, and less safety issues
without cost penalty

Solid tantalum capacitors have traditionally been popular in small hand held devices and in servers, but
recent polymer and ceramic capacitor improvements have made tantalum less attractive.
Raw material sourcing issues have also reduced tantalum’s market share (increased price).
Extreme care must be taken with tantalum capacitors to avoid reverse voltage or overvoltage, as these
conditions can cause the capacitor to burn out.
In spite of these concerns, tantalum capacitors are highly reliable and have been very popular in military
applications. However, modern ceramic capacitors can achieve nearly the same capacitance with lower
ESR and ESL and are replacing tantalums in many applications.
Only solid tantalum capacitors should be considered in modern applications. The "wet slug" type
tantalum capacitors are not suitable in today's designs.

4.16
Hardware Design Techniques

OS-CON (Sanyo)

‹ Offers large capacitance and very low ESR in relatively small package
with > 16V rating
‹ Useful for generating 100µF or more capacitance, and maintaining ESR of
< 50 mΩ (5 mΩ available)
‹ Can handle 0.5 to 4A of RMS ripple current
‹ Suitable for high current designs requiring very low voltage ripple
‹ > 5000 hours at 105°C with capacitance remaining within ±20%, with some
ESR increase
‹ Suitable for long lifetime products
‹ Lowest Height 4mm
z Not suitable for portable designs
z Used a great deal in servers, CPU voltage regulators
‹ Cost $0.50 to $1.00/100k piece price

OS-CON capacitors are attractive due to their reduced ESR and ESL. "OS-CON" is a trademark of
Sanyo and stands for "organic semiconductor."
OS-CON is an aluminum solid capacitor with high conductive polymer or organic semiconductor
electrolyte material. OS-CON achieves low Equivalent Series Resistance (ESR), excellent noise
reduction capability and frequency characteristics. In addition, OS-CON has a long life span, and its
ESR has little change even at low temperatures since the electrolyte is solid.
Other companies such as United Chemi-con have come out with similar products based on an aluminum
polymer technology that may have slightly lower lifetime but are more cost effective for the same
capacitance and similar ESR.

4.17
Hardware Design Techniques

POSCAP (Polymer Organic


Semiconductor Capacitor)

‹ Solid electrolytic capacitor with Sanyo-specific organic compound


‹ Offers reasonably low ESR (50mΩ down to 5mΩ)
‹ Small Size and Height
‹ Values up to 1000µF available
‹ Generally low voltage (up to 25V but limited capacitance)
‹ Cost 47µF, 6.3V: $0.40 to $1.50 at 100k piece price
z Less expensive than similar specified ceramic
z More expensive than similar standard electrolytic or tantalum

This capacitor type is similar to the OS-CON and also made by Sanyo.
"POSCAP" is a solid electrolytic chip capacitor. The anode is sintered tantalum, and the cathode is a
highly conductive polymer. "POSCAP" has a low ESR and excellent performance at high frequency
with a low profile and high capacitance. In addition, it has high reliability and high heat resistance.
Panasonic also has a “Special Polymer” cap that has similar specifications in the same footprint as the
"POSCAP." The cost is also comparable for similar values.

4.18
Hardware Design Techniques

Impedance of Various 100µF Capacitors


100
Aluminum Switching Type, 10V
Low ESR Tantalum, 10V
10 Polymer Tantalum, 4V
SP-Cap (SL Series), 2V
Ceramic, 6.3V ESL = 16nH
1
Z Ω

ESL = 1.6nH
100m

10m

1m
100 1k 10k 100k 1M 10M
FREQUENCY (Hz) ESR = 0
1 1
Self-Resonant Frequency =
2π ESL×C

This is a graph adapted from Panasonic's "S-P Capacitor Technical Guide" showcasing their “Special
Polymer” capacitor technology. It may cast a slightly unfavorable light on their competitor’s product
(namely Polymer Tantalum) but it shows the general trend of various capacitor types.
General purpose aluminum electrolytics have slightly higher ESR than the switching type. However,
they are not recommended for switching supply applications.
As you can see here, the ceramic capacitor has the lowest ESR at the switching frequency and the
lowest ESL.

4.19
Hardware Design Techniques

Ceramic Capacitors

Dielectric Granules Nickel Conductor


‹ Multilayer ceramic capacitors (MLCC) offer extremely low ESR (<10mΩ) and
ESL (<1nH) in a small thin package
‹ "0603" is 1.6mm (l) × 0.8mm (w) × 0.8mm (h)
‹ Best cost for performance from 1pF to 40µF
z Most suitable capacitor for portable power, decoupling supplies
z Available up to 100µF, but become larger and more expensive above 10µF
z Cost is dependent on value, with lower values being cheaper due to less
layers being printed. More popular values cheaper.
z 1µF 16V, 0603 is < $0.01 in 100k pieces
‹ Resilient to fault conditions – voltage surges etc.
‹ Capacitance value decreases with increasing voltage (stick with X7R types
for best results)

Due to low ESR, ceramic capacitors can handle very high ripple currents for their size. They are
exclusively used in compensation and small signal circuits, and are very good for reducing ripple voltage
on the input and output of switching supplies. Their ESL is also low, making them good choices at high
frequencies.
Due to their limitation in upper capacitance value, they are often paralleled with other capacitor types on
the input and output of switching supplies.
The ESL of a standard multilayer ceramic capacitor in an 0603 case is approximately 1 nH.
Other more expensive ceramic technologies are available which can reduce the ESL to less than 100 pH.
One type is the low inductance chip capacitor (LICC) which has a reverse-geometry where the
terminations are located on the long sides rather than the short sides of the rectangular package.
Other low inductance types are the inter-digitated capacitors (IDC), land grid array (LGA) capacitors,
and low-inductance capacitor arrays (LICA).
Details of these low inductance capacitors can be found at the AVX website, www.avxcorp.com, and in
the following reference:

Tim Sullivan, "Choosing Decoupling Capacitors," Electronic Products, December 2007, available at
www.electronicproducts.com/ShowPage.asp?FileName=farr_avx_dec2007.html.

4.20
Hardware Design Techniques

Which Ceramic Capacitor Provides


More Capacitance?
5.00
4.7µF, X7R,1206
CAPACITANCE (µF)

3.89µF @ 12V
4.14

3.3µF, X7R,1206
3.28 2.73µF @ 12V

2.42
4.7µF, X5R,0805 1.64µF @ 12V

1.56

0.748µF @ 25V
0.70 –84%
0 5 10 15 20 25
DC BIAS VOLTAGE (V)

‹ Generally more temperature stable parts (X7R) have less variation due to
applied voltage

A potential problem with ceramic capacitors is their capacitance change with dc bias voltage. This may
or may not be a problem in decoupling applications, but it can be a real issue if the capacitor is used in a
critical compensation network, for example. Another place where the voltage coefficient is a problem is
where the capacitor is used in the signal path, and the signal modulates the voltage. This can produce
unwanted harmonics because of the signal-dependent capacitance.
These plots show capacitance versus applied dc bias voltage for three types of ceramic capacitors:

4.7 µF, 25V, X7R, 1206 case style


4.7 µF, 25 V, X5R, 0805 case style (higher K, smaller case)
3.3 µF, 25 V, X7R, 1206 case style

It is interesting that the 3.3 µF (at zero bias) X7R capacitor actually has a larger capacitance at 12 V bias
than the 4.7 µF X5R type. At full rated voltage, the X5R capacitor is down 84% from its zero bias value.
Note that the X7R capacitors (which have a slightly lower dielectric constant) maintain their capacitance
better with dc bias voltage than the X5R type.
In addition, the X7R types perform better over temperature and should be used in preference to the X5R.
Avoid using Z5U and Y5V types which are worse.

4.21
Hardware Design Techniques

Inductor Basics
‹ An inductor is a magnetic energy storage element which
consists of a conducting coil surrounding a core, usually made
of ferrite material or powdered iron
‹ Air: µ = 1.257x10-6 H/m
λ ‹ Ferrite U M33: µ = 9.42x10-4 H/m
‹ Nickel: µ = 7.54x10-4 H/m
‹ Iron: µ = 6.28x10-3 H/m
‹ Ferrite T38: µ = 1.26x10-2 H/m
‹ Silicon GO steel: µ = 5.03x10-2 H/m
‹ Supermalloy: µ = 1.26 H/m

2
μ ⋅n ⋅ A
L(Henries) = ‹ Larger core cross-sectional area,
λ coil turns and/or core permeability
increase inductance
µ = permeability of core,
n = number of turns,
A = cross-sectional area of core,
λ = effective length of core

The practical inductor consists of a conductor coil wound on a ferromagnetic core. This combination
yields an inductance (L) that offers a reluctance to a change in current, and therefore the current through
an inductor cannot change instantaneously.
The rate of change of current through an inductor (di/dt) is determined by the inductance and the voltage
dropped across the inductor, given by the expression:
v = L×di/dt.
The use of ferromagnetic material as the inductor core allows energy to be stored in the inductor. When
a positive voltage is applied to the inductor, the current increases, and energy is added to the inductor
magnetic field.
The inductive impedance can be calculated from:

ZL = jωL
where:
j = √(–1)
and
ω = radian frequency = 2π × frequency in Hertz

4.22
Hardware Design Techniques

Wire and Strip Inductance Calculations

2R L, R in mm

WIRE INDUCTANCE = 0.0002L ln ( 2L


R
) - 0.75 μH

EXAMPLE: 1cm of 0.5mm o.d. wire has an inductance of 7.26nH


(2R = 0.5mm, L = 1cm)

W H

2L
( )
STRIP INDUCTANCE = 0.0002L ln W+H + 0.2235 ( W+H
L ) + 0.5
μH

EXAMPLE: 1cm of 0.25 mm PC track has an inductance of 9.59 nH


(H = 0.038mm, W = 0.25mm, L = 1cm)

A piece of wire or a PCB trace will have inductance, even though it is not a coil.
In addition to the resistance (basically a dc spec), a trace (wire or ground plane) will have a frequency
dependent impedance component (known as inductance).
Inductive impedance increases linearly with frequency. This can become significant at higher
frequencies.
The above figure shows simple equations to calculate the inductance of a length of wire or a rectangular
strip conductor. The assumption is that the return current path is a fairly long distance away from the
wire or strip—the wire or strip is treated as a circuit element.
However, on a PCB the actual inductance also depends on the loop area which includes the current
return path, and the above equations are no longer valid. There are modeling packages available to
determine the actual trace impedance in these cases.
For example, the effective inductance of a gate drive trace may be as much as a factor of 10 lower than
predicted by the above equation if it runs directly over a ground plane.

4.23
Hardware Design Techniques

Inductor Basic Current/Voltage Relationship


i
i

+
L di V
V =
dt L

0 t

1 2
E( joules) = ⋅L ⋅i
2
‹ Inductor current increases at rate of V/L
‹ If inductor current tries to change rapidly (switch opened), a large
voltage will be generated as the magnetic field collapses
‹ Energy stored is proportional to the inductance and the square of
the current.

You cannot change the current through an inductor instantaneously. When a voltage is applied to the
inductor, the current increases, and energy is added to the inductor magnetic field. When the switch is
opened, the inductor current goes to zero, and a large voltage is generated as the magnetic field
collapses.
The inductor magnetic field can only hold a finite amount of energy before the ferromagnetic material
will saturate. Saturation causes the inductance to decrease and ripple current to increase in a switching
supply. When making an inductor selection, it is important to check that the core saturation current (ISAT)
is greater than the application’s peak inductor current,

(IPEAK = IOUT + IRIPPLE/2).

4.24
Hardware Design Techniques

Inductor Model

ACTUAL MODEL
IDEAL (APPROXIMATE)

R
C
L CORE

Just as with the capacitor, a real inductor will have a more complex model.
The inductor will have series resistance (called DCR for dc resistance) just as in a capacitor. There is
also a small amount of parallel parasitic capacitance which can be relevant in RF applications.
Another consideration is the inductor self-resonant frequency. A practical example would be an inductor
of 10 µH which has an equivalent distributed capacitance of 5 pF. The self-resonant frequency can be
calculated as follows:

1 1
f resonance = = 22.5 MHz
2π LC

The switching frequency of the regulator should be at least ten times less than the inductor self-resonant
frequency. In most practical designs with switching frequencies less than 2 MHz this will be the case,
but a quick calculation is a good idea.

4.25
Hardware Design Techniques

Inductor Power Losses


LOSS FUNCTION OF
B

Magnetic Hysteresis ‹ Core Material


‹ Core Volume
H
‹ Flux Density
‹ Frequency

Eddy Currents ‹ Core Material


‹ Core Volume
‹ Flux Density
‹ Frequency

Winding Resistance ‹ Wire Size


‹ Number of Turns
‹ Core Volume

2πfL
‹ Figure of Merit: " Q" =
R

Another important consideration is that the temperature of the inductor will rise due to internal power
losses. The designer needs to consider winding resistance loss (sometimes referred to a copper loss) and
core losses from eddy currents (often referred to a core loss).
Winding resistance power loss is due to the effective current (IRMS) flowing through the resistance (RDC)
of the conductor winding, simply expressed as:
PCU = RDC×IRMS2.
Inductor data sheets typically specify a temperature rise at a given current—for example, the equivalent
dc current yielding a 40°C temperature rise. The lower value of the temperature rise and saturation
current is called the "rated" current of the inductor.
The mechanism of core loss is more complex. Recall that current flowing through an inductor winding
induces a magnetic flux in the ferromagnetic material, or the core. So the changing current in the power
inductor generates a changing flux density (BAC) and the reluctance of the core material tends to oppose
this BAC.
Just like the current flowing through the conductor results in copper losses due to the conductor's
resistance to the flow of current, the core's reluctance to a changing flux generates a core loss. The core
loss is determined by the type of core material, the amount of material, the BAC, and the frequency of
change, that is, the switching frequency (F). A good power inductor data sheet will simplify this
equation based on a calculated BAC for the operating condition of the inductor and the core material and
size.
While the data sheet may specify a temperature rise current, it is important to note that where the core
losses are significant, the inductor will reach the specified temperature rise at lower rms current because
of the additional temperature rise caused by the core loss.

4.26
Hardware Design Techniques

Inductor Functions in Switching Converters


BUCK BOOST
Input
Current
Input
Output
Output
Current
On-Time
Load On-Time Load
Current

Input
Input
Current
Output Output

Off-Time
Current Load Off-Time Load

VOUT ( VIN − VOUT ) VIN (VOUT − VIN )


ΔIL = ⋅ ΔIL = p-p Inductor Current ΔIL = ⋅
VIN fSW ⋅ L VOUT fSW ⋅ L

‹ Increasing fsw, reduces ripple current for same VIN/VOUT conditions


‹ Higher fsw = smaller value inductor = smaller physical inductor, and lower DCR
‹ However switching losses degrade efficiency at higher frequency
‹ Use ΔI ≈ 0.3 IDC (Buck Iload, Boost Iin) as starting point for inductance value
‹ Choosing L is a size/efficiency tradeoff

The primary function of an inductor in a switch mode power supply is to act as an energy storage device.
Another function of the inductor is to act as a filter—either on the input or output or both.
This figure reviews the basic operation of a buck and boost converter and how the inductor acts to step
down (buck) or step up (boost) the input voltage.
The calculation of the required inductor value begins by determining the maximum peak-to-peak ripple
current, ΔIL. This value is typically chosen to be about 30% of the output dc load current for a buck
converter, and 30% of the input dc current for a boost converter.
The inductor value itself can then be found by solving the above equations for L, given the value of ΔIL,
VIN, VOUT, and fSW.
Note that increasing the switching frequency reduces the value of the inductor required for the same
ripple current.
The ADIsimPower design tool calculates the proper value for the inductor based on the customer
requirements as well as determines the exact part and manufacturer for use in the bill of materials.
The ADIsimPower vendor database contains over 4000 individual parts. The inductor core loss data has
been included in the inductor model and is not generally published by the manufacturer.

4.27
Hardware Design Techniques

Inductor Specifications
3
DCR 2 SRF 4 5
Part L1 Percent max typ Isat Irms
number (µH) tol (Ohms) (MHz) (A) (A)

DO3316P-682_L_ 6.8 20,10 0.027 38 4.6 4.4


DO3316P-103_L_ 10 20,10 0.038 30 3.8 3.9
20

1. Inductances may vary 20 to 30% from the nominal inductor value! Current ripple
could be increased 43%. Coilcraft offers 10% and 20% tolerance.
2. DCR: DC resistance. Physically smaller inductors (for same L) will have larger
DCR due to reduced wire thickness and increased efficiency losses.
3. SRF: Resonant frequency of inductor and parasitic capacitance of inductor—
typically not a concern for design.
4. Isat: Inductor current level for which the inductance falls 10% using coilcraft, for
others it is –30%! You don’t want to operate at –30% (–10% is OK)
5. Irms (Also called Imax): At this RMS current level the temperature of the inductor
will rise 40°C. Inductors are usually rated to 125°C. Typically you don’t want to
have inductor rising much more than 40°C, because saturation current level is
further decreased.

This figure shows specifications for a particular inductor taken from the manufacturer's data sheet (in
this case, Coilcraft).
The inductance value can vary 20% to 30% from the nominal value, however, Coilcraft offers 10% and
20% tolerances. Since inductor ripple current is inversely proportional to L, inductances less than
nominal will cause an increase in ripple current.
Another important specification is the saturation current, Isat, which is the inductor current level for
which the inductance falls 10% (Coilcraft inductors). For other manufacturers, it is –30%.
Operation at 10% saturation is acceptable provided it represents the maximum load and worst case
conditions. Knowing this is valuable in selecting the right inductor value. Inductors which are larger
than necessary take up more real estate and are more expensive.
The specification Irms (or Imax) is the rms current level at which the temperature of the inductor will
rise 40°C. The maximum allowable temperature is usually 125°C. If you allow the inductor temperature
to rise more than 40°C the saturation current level will probably be further decreased.

4.28
Hardware Design Techniques

Inductor Cores and Their Characteristics

Ferrite Core
Saturates Quickly

Powdered Iron Core


Saturates Gradually

‹ Natural air gaps in powdered iron core cause a more gradual saturation curve for
same physical size.
‹ Powdered iron cores suited to applications requiring large instantaneous current
(i.e., camera flash, VRM power supplies).
‹ Powdered iron cores more expensive and have higher core losses.

There are two basic types of cores commonly used for inductors: powdered iron and solid ferrite.
The powdered iron core has natural air gaps which cause a more gradual saturation curve for the same
physical size (compared to solid ferrite cores). Because of the gradual saturation, powdered iron cores
are more suited to applications requiring large instantaneous currents such as camera flash and VRM
power supplies.
Ferrite core inductors saturate quickly but are lower cost and have lower core losses. They are therefore
more suited for standard POL switching supplies.

4.29
Hardware Design Techniques

Inductor Selection Considerations


‹ Benefits of low value inductors BUCK CONVERTER
z Lower DCR LOWER INDUCTOR VALUE
IL
z Higher saturation current FULL
z Higher di/dt LOAD
z Faster switching frequencies
z Improved transient response LIGHT
LOAD
z Less output capacitance required for given
transient performance (DISCONTINUOUS) t

‹ Benefits of high value inductors


BUCK CONVERTER
z Lower ripple current HIGHER INDUCTOR VALUE
IL
z Lower AC loss (skin effect, hysteresis) FULL
z Lower RMS current in switches LOAD
z Lower RMS capacitor current (mainly output)
z Continuous inductor current over wider load LIGHT
range LOAD
z Less capacitance required for an equivalent (CONTINUOUS) t
output ripple

This figure examines the tradeoffs between low and high value inductors. The ultimate choice depends
on the particular design.
The ADIsimPower design tool strives to achieve the desired performance with the lowest practical
inductor. This philosophy is the proper one for most modern portable systems.
On the other hand, higher value inductors do have certain advantages as shown in the figure. Under
certain conditions they may be the desirable choice.

4.30
Hardware Design Techniques

Saturation of Inductors
12 7
Current Limit
10 6

0.9L

Inductor Current (A)


5
8
L (µH)

0.7L 4
6
-40C 3
4
+85C 2
2
1
0
0
0 1 2 3 4 5 6 0 1 2 3 4 5
Current (Adc) Switc hing Cyc les

‹ The inductor core can become biased with too much current resulting in a sudden drop in
inductance.
‹ Operating with IDC + ΔI / 2 equal to a current ISAT which causes a 30% drop in L will cause
issues:
z Ripple current increase > 40%
z May hit current limit for lower DC current than desired
z Core saturation results in core efficiency loss
z Greater di / dt results in increased EMI leakage
‹ Over designing will result in physically large inductors
z 10% saturation for peak current at maximum load and worst case conditions is a
reasonable compromise. For the above example, this would be about 4A.
z Check Inductor saturation curves vs. temperature—Don’t just read spec!

This figure shows what happens to the inductor ripple current when the saturation current (Isat) is
exceeded. Exceeding the saturation current causes a sudden drop in inductance and a corresponding rise
in the inductor current.
The saturation current (Isat) is the current at which the inductance drops by a maximum of 10% below
the lower limit of its value specified at 0 A dc bias. This is the definition Coilcraft uses. Other
manufactures use different percentages.
The inductance at Isat is measured at the specified ambient temperature by applying dc bias for a short
period of time to minimize self-heating.
Selecting the inductor such that it will reach 10% saturation for peak current at maximum load and worst
case conditions is a good compromise. For the example shown in the figure, this would be about 4 A.

4.31
Hardware Design Techniques

Shielded or Unshielded?

‹ Shielded inductors reduce EMI and potential


interference
‹ RF designs almost always use shielded
inductors
‹ Manufacturing cost is higher
‹ Current rating per mm2 is reduced

Model L, µH DCR, Ω SRF, MHz ISAT, A IRMS, A Shielded Size, mm Price, $ 1k


DO5022P-103ML_ 10 0.031 30 10 4.3 No 18 x 15 0.73
DS5022P-103ML_ 10 0.04 30 8 3.9 Yes 18 x 15 1.11

To shield or not to shield, that is the question.


Shielding will reduce EMI. Although the cost will increase slightly, as is shown in the figure. Shielding
will tend to lower the saturation current, which in turn lowers the maximum allowable rms current.
Shielded inductors are very common in RF applications.
The above figure compares two similar inductors from Coilcraft. Both are the same size and value, but
the shielded version has slightly lower Isat and Irms currents.

4.32
Hardware Design Techniques

New Inductor Technology:


Multilayer Ceramic Inductors for Power
CROSS SECTION VIEW

Wire Coil Ferrite Core Ag Coil Ferrite Core

Standard Wire Coil Printed Ag (silver) Wire Coil

‹ Characteristics of Ceramic Inductors


z Low cost (about 50% to 65% of coil types with similar
specifications)
z Small size (2.5mm x 2.0mm x 1mm(h), 1.5µH, ISAT = 1.5A)
z Lower DCR improves efficiency
z Useful < 2.2μH (drives IC manufacturers to higher switching
frequencies)
z Imax = 1.5A due to small printed Ag (silver) wire coils
z Usable to about 1.0A due to saturation
‹ Manufacturers: FDK, Murata, TDK, Toko (in development)

Multilayer ceramic inductors lower cost, smaller size, lower DCR than standard wire coil inductors. The
only disadvantage is that the saturation current is typically limited to about 1.5 A because of the small
printed silver wire coils.
The upper range of inductance is limited to a few microhenries.

4.33
Hardware Design Techniques

Coupled inductors
‹ A coupled inductor is an inductor with multiple windings on the same
core
‹ Coupled inductors are basically transformers with poor coupling and low
magnetizing inductance
‹ A flyback transformer could be considered a coupled inductor with good
coupling

Lsp 1:1 Lss

Lm

Ideal Transformer Coupled Inductor Model

A coupled inductor is an inductor with multiple windings on the same core. It is similar to a transformer
but has more losses and a more complex model. It can be viewed as a transformer with poor coupling
and low magnetizing inductance.
This figure shows the relationship between the ideal transformer and a coupled inductor. The 1:1
transformer in the model is an ideal transformer (all it does is offer isolation)
The coupled inductor is more complex, since no simplifications can be made to the circuit model.
Coupled inductors generally have a 1:1 turns ratio, so Lsp = Lss.
Often Lm is of the same order of magnitude as Lsp and Lss.

4.34
Hardware Design Techniques

Calculation of Sheet Resistance


and Linear Resistance
R
ρZ
Z
R=
XY
ρ = RESISTIVITY
Y
X

SHEET RESISTANCE CALCULATION FOR


1 OZ. COPPER CONDUCTOR:
ρ = 1.724 X 10–6 Ω cm, Y = 0.0036cm

R = 0.491 Z mΩ
X
Z
= NUMBER OF SQUARES
X
R = SHEET RESISTANCE OF 1 SQUARE (Z = X)
= 0.491m Ω /SQUARE

We will now examine the resistor, another passive component.


All conductors have some resistance (at least when operating above 0°K).
Using large area ground planes decreases the resistance, but cannot eliminate it. And from ohm’s law we
know that a current flowing through a resistance will cause a voltage drop across the resistance.
The resistance of a trace (or a ground plane) can be calculated by taking the resistivity of the material,
which will typically be given in a resistance per unit volume (squares) of the conductor material, and
multiplying by the number of the squares.
In the above example, the sheet resistance of 1 oz. copper, which is a typical PC board material, is
calculated as 0.491 mΩ/square.

4.35
Hardware Design Techniques

Mismatched Resistor TCs Can


Induce Temperature-related Gain Errors
1 COMP PGATE 6
C2 R2
ADP1864 VIN
C1
2 GND CE1
IN 5
PGND
VFB RS
AGND
3 FB CS 4 D1
CE2
R1 R2 49.9kΩ, –100ppm/°C VOUT = +1.2V
U1 L1
100kΩ
+100ppm/°C

R1 Critical high dv/dt


VFB = VOUT or di/dt trace
R1 + R2
‹ R1/R2 divider sets output voltage (internal reference is 0.8V)
‹ Assume TC of R1 is +100ppm/°C and TC of R2 is –100ppm/°C
‹ Temperature change of 100°C causes gain error of +0.66%
‹ This translates into an error of +8mV at the buck output
‹ Avoid further mismatch errors by keeping R1 and R2 away from heat
generating components such as D1 and L1
‹ 1% surface mount chip resistors available in TCs to 25ppm/°C

As we have seen, processors with lower core voltages, such as FPGAs, are placing tighter tolerances on
power supply voltages. The accuracy of the feedback network in this circuit must be considered as part
of the overall error budget in the final output voltage. The following example illustrates the point.
For nominal resistor values R1 = 100kΩ, R2 = 49.9kΩ, gain = 0.6666666
Assume that R1 and R2 have temperature coefficients of +100 ppm/°C and –100 ppm/°C, respectively.
For a 100°C temperature change, R1 goes from 100 kΩ to 101 kΩ, and R2 goes from 49.9 kΩ to
49.4xkΩ. The new gain is 0.671096, which is 0.66% higher than the nominal gain. This becomes part of
the overall error budget which must also include the initial resistor ratio accuracy, line/load regulation,
reference accuracy, transient response, etc. Although 0.66% may seem small, the overall error budget
may be less than 5% for FPGA applications.
The use of 1% surface mount chip resistors for the gain setting resistors is recommended (such as
Vishay CRCW-series). TCs are available as low as 25 ppm/°C if required.
In the above diagram the red dotted line around the gate drive and feedback trace note that it should be
protected from interference for external signals.
Details of layout requirements regarding switching supplies follow later in this section.

4.36
Hardware Design Techniques

Current Sense Resistors Require Kelvin Sensing

SOLDER PAD SOLDER PAD

‹ If using a two pin/pad sense resistor, layout is critical to accuracy


‹ Kelvin sense tracks should be used at equal locations on each pad to
optimize voltage drop measurement
‹ It is also advisable not to place the sense resistor too close to heat
sources such as the MOSFET or catch diode in order to minimize
temperature rise

If the current sense resistance value is low, the lead resistance may be a significant source of error. To
compensate for this error it is common to use "Kelvin" (4-wire) connections, where the current flows
through one pair of leads and the voltage across the resistor is sensed with a second set in which no
significant current will flow. No current means no voltage drop. Such devices are, of course, more
complex and so more expensive.
This figure shows how careful board layout can often make them unnecessary. The current flowing
through the sense line to the VCC pin is small (assuming a switching controller with external switches),
and the voltage drop is usually insignificant. The load current flows through the sense resistor to the
load. If the VCC current is significant, then this sense trace must be made wide enough so that the voltage
drop is small with respect to the drop across the sense resistor.
Medium accuracy current sense resistors often use thick film or wire-wound technology, with resistance
accuracy of 1% to 5% and temperature coefficient of 20 ppm/°C to 80 ppm/°C. High precision sense
resistors use metal foil resistors and can meet specifications as high as 0.1% accuracy and 5 ppm/°C
temperature coefficient.
Sense resistors should not be placed close to heat sources such as MOSFET switches or catch diodes.

4.37
Hardware Design Techniques

Sense Resistor Considerations

‹ Choose maximum current limit, Imax


z Select the max operating current limit. This value can be obtained from the
data sheet of the controller
‹ Calculate sense resistor value
z Using the maximum sense voltage of the controller, the sense resistor can be
calculated as follows: Rsense = Vsense / Imax
‹ Calculate the sense resistor power
z The power is calculated simply as follows: P = Vsense × Imax
‹ Determine what accuracy you will require
z Once all this data is determined you can choose the sense resistor. Often it is
not possible to use just one resistor due to power and/or value constraints. In
these case multiple resistors are used in parallel / series
‹ Follow recommendations on the controller manufacturer's data sheet!!
‹ Sense resistors also used with hot swap controllers, etc.

This figure shows some considerations regarding the selection of the sense resistor.

4.38
Hardware Design Techniques

Active Components

4.39
Hardware Design Techniques

MOSFET Basics
N-CHANNEL MOSFET P-CHANNEL MOSFET
Drain Drain Drain Drain
= =
Gate Gate Gate Gate
Body
Body
Diode
Diode
Source Source Source
Source

‹ Three Terminal Device


z Gate
z Source
z Drain
‹ Two classical varieties: NMOS, PMOS
‹ Electrically isolated gate
‹ High input impedance
‹ Voltage-controlled device
‹ Internal parasitic body diode from drain to source

MOSFETs are the dominant switch technology for switch mode power supplies. Bipolar transitors are
sometimes used as switches, but suffer from a number of disadvantages compared to MOSFETs.
The figure shows two common schematic representations of the NMOS and PMOS FETs. The internal
parasitic body diode from drain to source is omitted in many schematics for simplification. The body
diode is unavoidable due to the way MOSFETS are constructed.
MOSFETs are easy to drive since they have high impedance inputs and are voltage-controlled devices.
The only current required is to charge/discharge the parasitic gate capacitance.

4.40
Hardware Design Techniques

N-Channel vs. P-Channel


VIN NMOS L VOUT VIN PMOS L VOUT

PWM PWM
CIN COUT CIN COUT
D D

VDRIVE

N-Channel MOSFET P-Channel MOSFET


‹ More difficult to drive ‹ Easier to drive
‹ Gate must be VIN + VGS(ON) to turn NMOS on ‹ Gate driven below VIN by at
‹ Drive voltage needed ≥ VIN + VGS(ON) least VGS(ON) to turn PMOS on
‹ Lower RDS(ON) ‹ Gate can be driven between
‹ Lower cost VIN and ground
‹ Can be run at 100% duty cycle
for portable applications
‹ Higher RDS(ON)
‹ Higher cost

This figure summarizes the basic differences between the N-Channel and the P-Channel MOSFET. Both
are shown in a typical asynchronous switching supply application.
N-Channel devices are more difficult to drive because the gate drive voltage (VDRIVE) has to be taken to
a more positive voltage than the source to turn on the device. If this voltage is not available it must be
generated, usually with a simple bootstrap circuit consisting of an external diode and a capacitor (the
current requirement is very low). P-channel devices require drive voltages which are less than the source
and therefore do not require an additional drive voltage supply. The gate of a P-Channel device can be
driven between VIN and ground.
P-Channel devices have intrinsically higher on resistance (RON) due to the deceased mobility of the
carriers in the P-type material. In addition, P-Channel devices are usually slightly more expensive than
N-channel devices.
MOSFET current handling capability can be increased by paralleling two or more similar devices. The
temperature coefficient of RON is positive which prevents current hogging. Diodes, on the other hand,
have a negative TC which makes paralleling difficult.
Parallel MOSFETs reduce the effective RON, but increase the gate capacitance and the resulting
switching losses.

4.41
Hardware Design Techniques

MOSFET Parameters of Interest


for Switching Converters
‹ Threshold voltage VGS(TH)
z Minimum gate bias which enables the formation of the channel
between source and drain
z Decreases with temperature
‹ On resistance RDS(ON)
z Total resistance between source and drain during on state
z Important parameter in determining current rating and power
dissipation
z Increases with temperature due to hole and electron mobility
decreasing with temperature
z Decreases with increased gate to source voltage
‹ RDS(ON) has positive TC (0.7%/°C to 1%/°C)
z Ideal for parallel operation
z Parallel MOSFETs tend to share current evenly
‹ Drain-source breakdown voltage BVDSS
z Maximum drain-to-source voltage device can endure without
avalanche breakdown in off state.

This figure summarizes the parameters of interest when selecting MOSFETs for switching converters.
The gate threshold voltage determines the drive voltage required to turn the MOSFET on and off. It is
important that this voltage does not exceed the input voltage. Otherwise, an additional supply will be
needed for the drive voltage.
The on resistance determines the current rating and the power dissipation of the MOSFET. It has a
positive temperature coefficient which allows multiple FETs to be paralleled without current hogging.
The drain-to-source breakdown voltage must include any transient voltage which may be generated due
to the inductor switching.

4.42
Hardware Design Techniques

Calculating Power in MOSFETs:


Conduction Loss
‹ The first of two primary mechanisms of BOOST
loss in a power FET is conduction loss VIN VOUT

‹ Conduction loss = (IRMS)2 × RDS(ON)

‹ RMS current in Boost MOSFET IPS RLOAD


z IPS(RMS) ≈ VIN √D × 1/((1–D)2 × RLOAD)
z D = tON/(tON + tOFF)
z VOUT/VIN = 1/(1–D)
BUCK
VIN IHS VOUT
‹ RMS current in Buck MOSFET
z IHS(RMS) ≈ IOUT × √D
z ILS(RMS) ≈ IOUT× √(1–D)
RLOAD
z VOUT/VIN = D
‹ Use RDS(ON) max value from ILS
data sheet at the appropriate
VGS drive voltage

When the FET is conducting it will be in series with an inductor for a period of the switching cycle.
Knowing that inductor’s dc current and ripple current values will allow you to calculate the rms current
through the FET during that cycle via the following equations.
Boost (CCM) (∆IL is defined as peak-to-peak inductor current) :

⎛ VIN ⎞ ⎡ ⎡ R LOAD ⎤ 1 ⎤
2
VIN ⋅ D
I PS( RMS) = ⎜ ⎟ D ⎢1 + ⎢ ΔI L (1 − D ) ⋅
2 ⎥ ≈
⎜ (1 − D) ⋅ R
2 ⎟ ⎢ ⎣ ⎥
⎝ LOAD ⎠ ⎣
2 ⋅ VIN ⎦ 3 ⎥
⎦ (1 − D )2 R LOAD
Buck (CCM) (∆IL is defined as peak-to-peak inductor current) :

⎡ 1 ⎛ ΔI ⎞ ⎤
2
I HS( RMS) = IOUT D ⎢1 + ⎜⎜ L ⎟ ⎥ ≈ IOUT ⋅ D
⎢ 3 ⎝ 2 ⋅ IOUT ⎟⎠ ⎥
⎣ ⎦

⎡ 1 ⎛ ΔI ⎞ ⎤
2
I LS( RMS) = IOUT (1 − D ) 1 + ⎜⎜
⎢ L ⎟ ⎥ ≈ IOUT ⋅ (1 − D )
⎢ 3 ⎝ 2 ⋅ IOUT ⎟⎠ ⎥
⎣ ⎦

For a first order approximation, use the RDS(ON) max value from the data sheet for the appropriate VGS
drive voltage. Be aware that RDS(ON) increases with temperature, so calculating power dissipation due to
conduction losses is actually an iterative process. However, when doing a “Will this FET work?” check,
the first order approximations shown above will likely answer that question for you.

4.43
Hardware Design Techniques

Calculating Power in MOSFETs:


Switching Loss
‹ The second of two primary mechanisms of loss in a power FET is
switching loss
‹ Goal is to switch between highest resistance state and lowest resistance
state as quickly as possible.
‹ Parasitic capacitances CGS and CGD slow down these transitions and thus
cause power dissipation proportional to frequency.

VDRIVE CGD = CRSS


CGS = CISS – CRSS
CDS = COSS – CRSS
CGD
RHI RG CDS
RLO
CGS

The second source of power loss in a power FET is switching loss. Ideally, the FET switches from its
highest resistance to its lowest resistance state in zero time. In practice, the finite switching time will
cause switching loss.
The parasitic capacitances CGS and CGD slow down the transition times and cause power dissipation
associated with the switching frequency.
As a reminder:

CGD = CRSS
CGS = CISS – CRSS
CDS = COSS – CRSS

Note that the parasitic capacitances are voltage dependent. Therefore the values in the table of the data
sheet are not very helpful. Use the capacitance versus VDS curves on data sheet to determine the value of
capacitance to use.

4.44
Hardware Design Techniques

MOSFET Switching Loss (Continued)


‹ The product of the current through and voltage across the inductor is a
triangle wave shape. The area of this triangle is the switching transition
loss.
‹ PSW ≈ ( VIN × IOUT ) × FSW × (tRISE + tFALL)/2

TEST CIRCUIT
Volts VDS VGS
VDD or Amps
IDS
ID
IG
VGS, Miller

SW t0

SWITCH OPENED AT t0 t1 t2

tRISE

This figure represents the turn on waveforms of a MOSFET.


Time period t1 is primarily due to the CGS capacitance and current drive capability of the driver.
Time period t2 is the main component in slowing switching and transitions and increasing power loss.
t2 is primarily due to the CGD or Miller capacitance and the current drive capability of the driver.
During switching transitions (time periods t1 and t2) there is both voltage across drain to source and
current flowing from drain to source. The presence of both current and voltage suggests there will be
power dissipation in the device.
The turn off waveforms follow the same signatures. The falling waveforms would be rising and the
rising waveforms falling in the turn off transition.
These losses are a function of how often the transitions occur, so switching losses increase with
switching frequency.
Many other factors take this analysis beyond 1st order approximations seen here. MOSFET
transconductance, reverse recovery of synchronous FETs, parasitic inductances.
IGT1 = VDRIVE – 0.5 × (VGS,Miller + VTH) / (RHI + RGATE)
IGT2 = (VDRIVE – VGS,Miller) / (RHI + RGATE)
t1 = CISS × (VGS,Miller – VTH)/ IGT1
t2 = CRSS × V DS(OFF)/IGT2
tRISE = t1 + t2
IGT3 = VGS,Miller / (RLO + RGATE)
tFALL = Q G(SW) / IGT3
Fsw= switching frequency

4.45
Hardware Design Techniques

Is MOSFET Power Dissipation Too High?


‹ The total loss (nearly) in the FET is the sum of the switching loss and
conduction loss
z PTOTAL = PCONDUCTION + PSWITCH
‹ Now that we have the total loss, we can determine if it is too high for thermal
stability. The answer will be (as with most things) “it depends.”
‹ MOSFET manufacturers will specify a maximum junction temperature (TJMAX)
as well as a junction to ambient thermal resistance (θJA) on their data sheets.
‹ One can roughly calculate the approximate junction temperature (TJ)
knowing θJA, estimated ambient temperature (TA), and total power (PTOTAL) as
follows:
z TJ = TA + θJA × PTOTAL
z If TJ > TJMAX, the FET you selected will not work. (for long)
‹ Because MOSFET RDS(ON) has a positive TC (0.7%/°C to 1%/°C), parallel
MOSFETs share current fairly evenly, so parallel connections are often used
in lieu of larger single FETs. Dual FETs are often connected in parallel, for
example

It should be noted that ӨJA is generally specified with a given type and area of copper. Chances are your
design may have different copper and area. Additional analysis must be done to get a precise TJ.
However, this 1st order approximation will likely answer the “Will this FET work?” question. There is
also a linear derating factor associated with increased ambient temperature. If the design pushes the
upper thermal limit, it might or might not work. Proceed with caution.
The failure mechanism in the FET is the heat itself. There is no inherent limit to the amount of current a
FET can pass, so if you were running a FET with a good heat sink, the amount of power it dissipated
could be significantly increased.
As indicated previously, MOSFETs can be paralleled in order to increase the total current capability at
the expense of more drive current.
See the thermal design discussion later in this section.

4.46
Hardware Design Techniques

Snubber Selection for Buck Converters


‹ Not generally required for <3A applications with good layout
‹ Add series RC to each switching FET
‹ Check each switch node
‹ Design will have less EMI and less component stress
‹ Gate resistance can control rise / fall time
SWITCH SWITCH NODE
NODE SNUBBER VOLTAGE
RISE / FALL TIME
CONTROL
RESISTOR LPARASITIC
RSNUBBER

f1 = RESONANT FREQUENCY
(~50MHz to 200MHz)
CP1 MEASURED WITHOUT SNUBBER
CP2

CP1 ≈ COSS
1
RESONANT FREQUENCY = f1 =
2π LPARASITIC ⋅ CP1

A snubber is sometimes required to protect the FET from overshoot, undershoot, and ringing at the
switch node. This is caused by fast edges exciting the switch node capacitance and parasitic
inductances of parts and traces. Most of the time we do not care much about this unless the spikes
exceed device breakdown ratings, or the resulting EMI exceeds limits.
There is no good rule for the use of snubbers, other than use them when needed. If layout is bad, a
snubber may improve EMI at most any current; but with good layout we seldom use snubbers in buck
regulators under 3 A . As the current rises to >10 A snubbers are often used.
Snubbers are good at damping the ringing, but they consume power and have little effect on initial
overshoot. Rise time and fall time control can be used to limit this spike—sometimes a gate resistor is
used—but this method is lossy. Both methods reduce overall efficiency. A conservative approach is to
initially lay out the PCB with a place for a snubber network also rise time limiting resistors, realizing
that these may not be required after testing.
The figure shows an approximate equivalent circuit for designing the snubber network. When the FET
current changes rapidly, ringing can occur at the resonant frequency of the FET output capacitance
(COSS) and the parasitic inductance in the loop (LPARASITIC). The frequency of this ringing is generally
between 50 MHz and 200 MHz depending on circuit conditions.
Assume a resonant frequency of 100 MHz, and COSS = 300 pF, then LPARASITIC = 8 nH.

4.47
Hardware Design Techniques

Snubber Selection for Buck Converters


‹ Apply Load to a Buck converter and measure resonant frequency (f1) on switch
node transition
1
f1 =
2π • CP1 • LP
‹ Add a reasonable amount of capacitance (CP2) whose value is somewhere between
COSS of the low side FET and 2×COSS to the switch node to ground and measure the
resonant frequency again (f2).
1
f2 =
2π • (CP1 + CP2 ) • LP
‹ The result is two equations and two unknowns (CP1 and LP)
‹ The next step is to find an RS value which will slightly over-damp the LRC resonance
by setting Q roughly equal to 0.9

Q = LP /(CP1 + CP2 ) / R S
‹ The power dissipated in the snubber resistor will be ½ CV2 for turn on and ½ CV2 for
turn off:

P = CP2 • VIN2 • fSWITCHING

This is the first order approximation process for finding snubber values for a buck converter. Proper
measurement techniques to reduce parasitics are necessary when measurement switch node resonant
frequencies.
The first step is to measure the ringing frequency on the switch node, f1. Then add an external capacitor
from the switch node to ground, CP2, with a value between COSS (from the FET data sheet) and 2⋅COSS.
Measure the new resonant frequency, f2.
The resulting two equations can then be solved for the two unknowns, LP and CP1.
The next step is to determine the approximate value of RS which will slightly over-damp the LRC
resonance by setting Q equal to 0.9. The equation for Q can then be solved for RS.
One can see that the added capacitance is directly proportional to the total loss in the snubber. There
may be an iterative process required to find a value that minimizes losses and provides adequate
damping of the ringing on the switch node.
RS and CP2 can be further optimized by experimentation—increase CP2 slightly and decrease RS.
Snubbers are more effective the closer you can physically get them to the die voltage which you trying
to snub, so the RC should be placed as close to the drain-to-source on the low side FET in a buck as
possible. Make sure to adequately size the resistor to handle the power dissipation. The capacitor will
not dissipate any notable amount power.
The power dissipated in the snubber resistor is approximately
P = CP2⋅VINPUT2⋅FSWITCHING.
For an example power calculation, assume CP2 = 600 pF, VINPUT = 12 V, FSWITCHING = 1 MHz, then P =
86 mW.

4.48
Hardware Design Techniques

Effect of Snubber on the Switch Node


for a Buck Converter
NO SNUBBER WITH SNUBBER

≈ 65 MHz RINGING

VERTICAL: 5V/div VERTICAL: 5V/div


HORIZONTAL: 20ns/div HORIZONTAL: 20ns/div

Here we show the effect of the snubber on the switched waveform. Note the reduction (although not
elimination) of the initial overshoot and ringing.

4.49
Hardware Design Techniques

Diode Rectifiers
‹ Standard Recovery Silicon (not recommended)
‹ Fast Recovery Silicon (not recommended)
z 50V to 1000V breakdown
z 100ns to 500ns recovery time
‹ "Ultrafast" Recover Silicon (only for high voltage
applications)
z 50V to 1000V breakdown
z 20ns to 75ns recovery time
‹ Standard Schottky
z 20V to 100V breakdown
z Switcher "Workhorse"
z 0ns recovery time
‹ Low Forward Voltage Schottky
z 10V to 40V breakdown
z 0ns recovery time

This figure lists diode rectifiers starting with the oldest technology, standard recovery silicon diodes.
These were the first type of rectifiers, such as the 1N4004 and 1N5404. Reverse breakdowns were high,
but recovery time was not specified.
The addition of gold doping yielded "fast recovery" silicon diodes of 100 ns to 500 ns recovery, such as
the 1N4936.
These were followed by "ultrafast" recovery silicon diodes with 20 ns to 75 ns recovery time and up to
1000 V reverse breakdown, such as the MUR140 and ES1D.
Today, the standard Schottky diode is the workhorse diode of asynchronous switchers. Certain varieties
are available with breakdowns up to 200 V.
A variety of Schottky diodes are optimized for low forward voltage drop with breakdowns up to 40 V.
None of the Schottky diodes have measurable reverse recovery time.
Modern switching regulators use Schottky diodes in 99% of applications rather than the older silicon
varieties.

4.50
Hardware Design Techniques

Diode Parallel and Series Connections


I1
D1 D1 D2
I

D2

I2

‹ Diodes have a negative TC, therefore one will "hog" most of the current in
parallel connection. Use synchronous rectification.
‹ This may be useful in reducing overall parasitic inductance at the
expense of additional capacitance. Cannot assume currents will share
equally, however. Each diode must be rated for full current load.
‹ Series connection of diodes to increase effective breakdown voltage is
risky and should be avoided.

The power dissipated in the diode in an asynchronous switching converter often becomes a limiting
factor at high currents. Unfortunately, diodes can't be paralleled to share the current, because they have a
negative forward voltage temperature coefficient.
With two diodes in parallel, one will always have a slightly lower voltage drop, and this diode will carry
slightly more current than the other one, the voltage drop will decrease, it will carry even more current,
and eventually it will carry all the current.
Parallel diodes can be useful in reducing the overall parasitic inductance (at the expense of increased
capacitance) but will not work in current-sharing applications.
It is not advisable to connect diodes in series to increase the effective reverse breakdown.
The basic point here is that it is not advisable to try to use multiple diodes in place of one correctly sized
diode in either the parallel or series connection.

4.51
Hardware Design Techniques

Power Supply Layout


and Grounding

4.52
Hardware Design Techniques

Parasitics to Deal with in PCB Layout


‹ Trace Resistance
z DC and AC errors due to voltage drop
z Sheet resistance of 1 ounce copper = 0.491mΩ / square

‹ Board Capacitance
z Coupling into high impedances and noise-sensitive circuits
z Coupling between planes and to component pads
z Trace capacitance = 2.8 pF/cm2 for 1.5mm glass epoxy (εr = 4.7)

‹ Wiring Inductance
z Especially in low impedance circuits and filters
z Use wide conductors and ground planes to minimize

‹ Magnetic Coupling
z Inductor-to-inductor, especially toroids. Consider alternate mounting
directions
z Loop-to-loop, minimize loop areas, use ground planes

Besides the components loaded on a PC board, there are also many others parasitic elements that
contribute to the performance of the circuit.
Trace resistance of the copper is approximately 0.491 mΩ / square for 1 ounce copper. For 2 ounce
copper, it is one-half this value, or 0.246 mΩ / square, etc.
Trace capacitance is about 2.8 pF/cm2 for 1.5 mm thick glass epoxy (εr = 4.7).
Trace inductance is given by the equation on p. 4.23 in this book.
Magnetic coupling between inductors, especially toroids, is also a consideration. This effect can be
reduced by alternate mounting directions, minimizing loop areas, and the use of ground planes.

4.53
Hardware Design Techniques

Printed Circuit Board Resistance

Copper Thickness Resistance Coefficient, Reference 0.1


milliohms/inch/w inch wide trace,
(trace width w in milliohms/inch
inches)

2 0.983/w 9.83
1/2 oz/ft
2 0.491/w 4.91
1 oz/ft
2 0.246/w 2.46
2 oz/ft
2 0.163/w 1.63
3 oz/ft

Sheet resistance of 1 ounce copper = 0.491mΩ/square

Common values for resistance of traces for various weights of copper commonly used in PCB
manufacture.
Note that not all layers of a multilayer board need to be the same weight of copper. Thinner copper can
be used for signal layers, especially when using fine pitch components which require very narrow traces.
Power planes can benefit from heavier copper by lowering trace resistance and helping dissipate more
heat.

4.54
Hardware Design Techniques

Relative Dielectric Constant, εr ,for Common


PCB Materials
Test Frequency

Laminate 1kHz 1MHz


FR4 Glass Epoxy 4.4 4.7
FR5 Glass Epoxy 4.4 4.7
G9 Glass Melamine 7.2 7.5
G3 Glass Phenolic 5.5
G7 Glass Silicon 4.2 4.7
Epoxy Thermount 3.9
BT Epoxy 4.1
Epoxy/Polymide 4.4
Cyanide Ester 3.5

This table shows the approximate dielectric constants of a variety of materials used in PCB dielectrics.
These are rough averages of numbers from several handbooks and various manufacturers' literature,
which often disagree significantly. Most are for 25°C ambient conditions, but some are measured at
20°C.
If this is an issue for your design, it is best to check with the PCB manufacturer.

4.55
Hardware Design Techniques

Kirchoff’s Law Helps Analyze Voltage Drops


Around a Complete Circuit

POWER
I
LOAD
SOURCE

RL I

GROUND RETURN CURRENT


G1 G2
I

AT ANY POINT IN A CIRCUIT


THE ALGEBRAIC SUM OF THE CURRENTS IS ZERO
OR
WHAT GOES OUT MUST COME BACK
WHICH LEADS TO THE CONCLUSION THAT
ALL VOLTAGES ARE DIFFERENTIAL
(EVEN IF THEY’RE GROUNDED)

One of the biggest problems in system design is how to handle grounding. There are several competing
requirements that are dependent on the frequency and system complexity.
Unfortunately, there is no magic “cookbook” approach to grounding that will always guarantee success.
What we will do here is present some of the effects that must be considered when designing the system.
The main thing is to look at how and where the dc and ac currents flow in a PCB.
We reviewed the procedures for proper mixed-signal circuit grounding in Section 3. In this section we
will look at grounding issues associated with switching power supplies.
When we draw the ground symbol on a schematic, we assume that all ground points are at the same
potential. This is rarely the case, unfortunately.
Historically, "ground" was the reference level with which we measured various voltage levels in the
circuit. However, ground has also become the power return not only for digital signals but for analog
signals as well.
All signals that flow in a circuit must have a return path to complete the loop. Often we consider the
forward path only, but there always must be a return to close the loop or current cannot flow. This return
path is often through the ground plane.

4.56
Hardware Design Techniques

A More Realistic View of the Impedance


Between Grounds

LOAD
POWER
SOURCE
I

ΔV = VOLTAGE DIFFERENTIAL
DUE TO SIGNAL CURRENT AND/OR
EXTERNAL CURRENT FLOWING IN
GROUND IMPEDANCE

G1 G2
IEXT

The connection between two points on the ground plane is never zero impedance. There is always some
resistance and inductance, even in a large area heavy ground plane.
The magnitude of the impedance may be small, but it is not zero. And a current flowing through an
impedance causes a voltage drop.
This means that the two grounds in the diagram above will not be at the same potential.
It is important to consider the inductance of the ground as well as the resistance, especially as the
frequency increases.

4.57
Hardware Design Techniques

Switching Currents Flowing in Analog Return


Path Create Error Voltages
ID

IA
INCORRECT
+ +
VD VA ANALOG SWITCHING
VIN CIRCUITS CIRCUITS

GND IA + ID ID
REF
ID

IA CORRECT
+ +
VD VA ANALOG SWITCHING
VIN CIRCUITS CIRCUITS

GND IA
REF
ID

Because ground is the power return for all digital circuits, as well as many analog circuits, one of the
most basic design philosophies is to separate digital ground returns from analog ground returns.
If the grounds are not separated, not only does the return from the analog circuitry flow through the
analog ground impedance, but the digital ground current also flows through the analog ground, and the
digital ground current is typically much greater than the analog ground current.
As the frequency of digital circuits increases, the noise generated on the ground increases dramatically.
TTL and CMOS logic families are of the saturating types. This means that the logic transitions cause
large transient currents on the power supply and ground. CMOS outputs basically connect the power to
ground through a low impedance during the logic transitions.
And it’s not just the basic clock rate that is a problem. Digital logic waveforms are basically rectangular
waves, which implies many higher frequency harmonic components.
The same holds true for switching power supplies.

4.58
Hardware Design Techniques

Magnetic Field Lines and Inductive Loop


(Right Hand Rule)

I FLUX LINES
OPPOSING AROUND
OUTSIDE TOP WIRE
LOOP

REINFORCING
INSIDE
LOOP

FLUX
LINES
AROUND
BOTTOM OPPOSING I
WIRE OUTSIDE
LOOP

The right hand rule is useful in predicting the direction of the magnetic field lines produced by a current
flowing in a conductor.
If you point the fingers of your right hand in the direction of the flux density, the induced signal will
flow in the direction that your thumb is pointing.

4.59
Hardware Design Techniques

Magnetic Field Passing Through a


Loop Area Creates Magnetic Flux

MAGNETIC FIELD
INTO THE PAGE

+
VDC – MAGNETIC FLUX
IS A MAGNETIC FIELD
PASSING THROUGH
LOOP A LOOP AREA
AREA

A loop of wire carrying current is essentially an electromagnet whose field strength is proportional to the
current. Magnetic flux is proportional to the magnetic field passing through the loop area,
Magnetic Flux ∝ Magnetic Field × Loop Area
or more precisely,
ΦB = BA cosφ
Where the magnetic flux, ΦB, is the magnetic field, B, passing through a surface loop area, A, at an
angle, φ, to the area’s unit vector.
A look at the figure gives meaning to the magnetic flux associated with an electric current. A voltage
source pushes current through a resistor and around a loop of wire. This current generates a magnetic
field which encircles the wire. To relate the different quantities, think of grabbing the wire with your
right hand (applying the right-hand rule). If you point your thumb in the direction of current flow, your
fingers will wrap around the wire in the direction of the magnetic field lines. As those field lines pass
through the loop, their product is magnetic flux, directed in this case into the page.
Change either the magnetic field strength or the loop area, and the magnetic flux will change. As the
flux changes, a voltage is induced in the wire, proportional to the rate of change of the flux, dΦB/dt.
Notice that either a fixed loop and changing current or a constant current and a changing loop area—or
both—will the change the flux.

4.60
Hardware Design Techniques

Effects of Opening a Switch


+ –

OPEN THE SWITCH AND


+ MAGNETIC FLUX GOES TO ZERO,
VDC –
AND VOLTAGE IS INDUCED
EVERYWHERE ALONG THE WIRE

+ –
INDUCED VOLTAGE
(GROUND BOUNCE)

Suppose, for example, that the switch in the figure is suddenly opened. When current stops flowing, the
magnetic flux collapses, which induces a momentarily large voltage everywhere along the wire.
If part of the wire is a ground return lead, voltage that is supposed to be at ground will spike, thus
producing false signals in any circuitry using it as a ground reference.
Generally, voltage drops in printed-circuit-board sheet resistance are not a major source of ground
bounce. 1-oz copper has a resistivity of about 500 mΩ/square, so a 1 A change in current produces a
bounce of 500 mV/square—a problem only for thin, long, or daisy-chained grounds, or precision
electronics.

4.61
Hardware Design Techniques

The Effects of Switching on Loop


Area for Buck Converters
(A) IDEAL LEAD
SWITCHES INDUCTANCE VOUT
1

1 2
+ 2 1 2
VIN CIN COUT

GND
+ –
CHANGING GROUND
BOUNCE CASE 1: CURRENT LOOPS ARE RED
LOOP AREA
CASE 2: CURRENT LOOPS ARE BLUE
LEAD
HIGH-SIDE
(B) REAL INDUCTANCE VOUT
SWITCH
SWITCHES
1
+ 2 1 2
VIN CIN
– COUT
LOW-SIDE
SWITCH
GND
CHANGING + GROUND –
LOOP AREA BOUNCE

The best way to reduce ground bounce in a switching dc-to-dc converter is to control changes in
magnetic flux—by minimizing both current loop areas and changes in loop area. This is illustrated in
(A) which uses ideal switches. Although the input and output currents are roughly constant, as the
switch moves from Position 1 to Position 2, the total loop area rapidly changes in the middle portion of
the circuit. That change means a rapid change in magnetic flux, which in turn induces ground bounce
along the return wire. In some cases, as in the figure, the current remains constant, but the switching
produces a change of loop area, hence a change of flux. In (A) an ideal voltage source is connected by
ideal wires to an ideal current source (the inductor represents the ideal current source). Current flows in
a loop that includes a ground return. When the switch changes position, the same current flows in a
different path. The current source is dc and does not change, but loop area does change. The change in
loop area means a change in magnetic flux, so voltage is induced. Since a ground return is part of that
changing loop, its voltage will bounce. The circuit in (B) shows the same principle implemented with
real switches.
The fact that a change in magnetic flux will induce voltage everywhere along a ground return brings up
the interesting question: where is true ground? Because ground bounce means a voltage on the ground
return trace is bouncing with respect to some ideal point called ground, that point needs to be identified.
In the case of power-regulating circuits, true ground needs to be at the low end of the load. After all, a
dc-to-dc converter’s purpose is to deliver quality voltage and current to the load. All other points along
the current return are not ground, just part of the ground return, subject to losses due to inductance and
resistance.
In the following discussions, we will ignore the loop composed of VIN and the lead inductance, because
the current through this loop remains constant for each cycle.

4.62
Hardware Design Techniques

Buck Converter Layout


HIGH-SIDE
VIN SWITCH VOUT

1 2
CIN COUT
LOW-SIDE
SWITCH
GND
+ GROUND –
BOUNCE
CHANGING
LOOP AREA
(MINIMIZE)

VIN VOUT

COUT
CIN

PGND

This shows how the currents and loop areas change in a buck converter. A good way to minimize the
ground bounce is to minimize the changing loop area by careful placement of the components, primarily
the CIN capacitor. The critical loop is comprised of the high-side switch, the diode, and CIN. Capacitor
CIN bypasses the top of the high-side switch directly to the bottom of the low side switch. thereby
shrinking the changing loop area and isolating it from the ground return. From the bottom of VIN to the
bottom of the load, there is only a small loop-area change from one case to the next. Consequently, the
ground return bounce is minimized.
As shown in the lower diagram, both CIN, COUT, and the diode should be connected together with short
leads to the power ground, designated "PGND" and shown by a triangle comprised of line segments
rather than a solid triangle.

4.63
Hardware Design Techniques

Boost Converter Layout


HIGH-SIDE
SWITCH
VIN VOUT

2 1
CIN LOW-SIDE COUT
SWITCH
CHANGING
GND LOOP AREA
GROUND (MINIMIZE)
+ –
BOUNCE

VIN VOUT

CIN COUT
PGND

A boost converter is essentially a reflection of a buck converter, so it is the output capacitor that must be
placed between the top of the high-side switch and the bottom of the low-side switch to minimize the
change in loop area. Here the critical loop to be minimized is comprised of the low-side switch, the
diode, and COUT.
Note that CIN, COUT, and the bottom of the low-side switch are connected through short leads to the
PGND.
This figure and the previous one illustrate the basic principle of identifying the critical loops and
minimizing their areas. This will now be applied to several actual SMPS layouts.

4.64
Hardware Design Techniques

Synchronous Buck Converter Layout


HIGH-SIDE
VIN SWITCH VOUT

CIN LOW-SIDE COUT


SWITCH
GND

CIN, 1210 TYPE

INPUT GROUND
und Input VIN TO HIGH-SIDE
TO LOW-SIDE
ow Side
DRAIN
SOURCEce

LOW-SIDE HIGH-SIDE
MOSFET MOSFET

SWITCH NODE
CONNECTS TO INDUCTOR

This figure shows the layout of the critical loop area in a synchronous buck converter.
Although the inductor has "continuous" high current, this current is switched alternately thru the top
(control) and bottom (synchronous rectifier) FETs. The current waveform in each FET is a pulse with
very high di/dt. This high di/dt flows in alternating directions in a loop comprised by the two FETs and
the input bypass capacitor. Any inductance in this loop causes voltage spikes on the switch node and
high-side drain with respect to the low side source, which can result in a variety of problems. To
minimize this inductance, the current loop between the input bypass capacitor and thru the two FETs
should be as short, fat, and tight as possible.
This example illustrates one way to get low inductance in the input ac current loop. Here, the SO-8 FETs
are counter-rotated to allow shorter connections to the high current paths. The high-side drain and low-
side source fit closely against the ceramic input bypass. The switch node connections fit to one compact
power plane. This arrangement effectively minimizes the critical loop area composed of the switches
and the input capacitor.

4.65
Hardware Design Techniques

ADP1821 Synchronous Buck Layout

VIN
ADP1821 CIN

PGND

VOUT

COUT

PVCC

AGND

After minimizing inductance and impedance in the critical high current power paths, the second key
concept is to prevent power and gate drive noise sources from interfering with the sensitive analog
connections. You want to keep noise current out of traces that are noise sensitive; some of the
approaches to doing this are based on Kelvin voltage sensing
ADI's design of controllers such as the ADP1821 allows significant noise voltage differential between
PGND (current return path for the gate driver) and AGND (ground reference for analog circuitry). This
permits a useful degree of separation of these two, which makes it easier to accomplish the desired
isolation. Some competitor's ICs do not allow this separation and the layout can suffer as a result.
This figure shows schematically where the critical loop is located. All connections to the PGND island
should be as short as possible. The PGND island is connected to the AGND island using a separate
trace. The feedback resistors, soft-start capacitor, and the feedback compensation network should be
connected to the AGND island.
Note that the output voltage is Kelvin sensed and connected to the feedback network using a separate
trace. This trace should be isolated from any noisy traces.

4.66
Hardware Design Techniques

ADP1821 Synchronous Buck Layout Details-1

This shows details of the layout of the critical components in the ADP1821 synchronous buck layout.
Note that CIN and COUT are both grounded to the PGND island. The high-side and low-side FETs are
rotated as previously described to minimize the critical loop area.
PGND connects directly to PGND foil. PVCC is bypassed to PGND foil with a short trace to a capacitor
very close to the chip.
The AGND island is located under the ADP1821 and is connected to the PGND by a separate trace.
This trace carries very little current. The FB divider, VCC bypass, soft-start capacitor, and control signals
are grounded to the AGND island.
This arrangement provides the most accurate ground return sense for the converter. Place components
close to the IC and make connections as short as possible.
This rotation of the ADP1821 allows direct connections from the driver outputs to the FET gates. Neat,
huh?
The CSL resistor connection to switch node is separate from that of SW pin for best accuracy in sensing
VDS on the low-side FET.
Short, fat power ground foil connects ADP1821 PGND to source leads of low side FET thru multiple
vias. This provides a low impedance ground return path for gate drive current. This path is separate from
analog ground.

4.67
Hardware Design Techniques

ADP1821 Synchronous Buck Layout Details-2

Do not run sensitive traces (such as FB, Soft Start, and Compensation) close to and parallel with noise
generators such as gate drive (DH and DL) and the switch node. Parallel traces encourage inductive
coupling, while mutual surface area can encourage capacitive noise coupling. Close proximity can make
both worse. Where additional copper layers are available, a ground plane (placed between noise
generating and noise sensitive nodes) can provide significant and helpful decoupling.
Gate drive traces (DH and DL) handle high di/dt so tend to produce noise and ringing. It is imperative
that they should be as short and direct as possible. If at all possible, avoid using feedthru vias in the gate
drive traces. If vias are needed, it is best to use two relatively large ones in parallel to reduce the peak
current density and the current in each via.
The switch node connects the source of the high-side FET to the drain of the low-side FET and the
inductor. This is the noisiest node in the switcher circuit with large ac and dc voltage and current (high
dv/dt and di/dt). This node should be wide to keep resistive voltage drop down. But to minimize the
generation of capacitively coupled noise, the total area should be small. The best layout will generally
place the FETs and inductor all close together on a small copper plane in order to minimize series
resistance and keep the copper area small.

4.68
Hardware Design Techniques

Asynchronous Buck Switching Controller Layout


1 COMP PGATE 6
C2 R2
ADP1864 VIN
C1
2 GND CE1
IN 5
PGND
RS
AGND
D1
3 FB CS 4
RBOTTOM CE2
VOUT
RTOP U1 L1

PGND
VIN

VOUT

AGND

FB TAP
AGND NOISY CONNECTION ON TOP OF BOARD
FROM VOUT

Here we show another example of a double-sided PCB layout for a switching supply. This is the layout
of the evaluation board for the ADP1864 asynchronous buck switching controller.
This layout follows the guidelines previously presented. The critical loop area to be minimized is
comprised of the high-side FET (U1), the current sense resistor (RS), the input capacitor (CE1), and the
diode (D1).
Note that high dv/dt and di/dt traces and the changing loop area have been minimized.
Ground islands are used for PGND and AGND.
The output voltage is sensed and routed to the feedback network using a separate trace on the bottom
layer of the board.

4.69
Hardware Design Techniques

Details of ADP1864 Double-Sided Board Layout


PGND
VIN

TOP LAYER

VOUT

AGND

FB TAP AGND NOISY CONNECTION ON TOP OF BOARD

BOTTOM LAYER
(SEEN FROM TOP) PFET GATE CONNECTION

FEEDBACK PGND
CONNECTION

PGND TO AGND
CONNECTION

This figure shows the top and bottom layers of the PCB in order to see the feedback connection, and the
PGND to AGND connection.

4.70
Hardware Design Techniques

Return Current Takes the Path of Least


Impedance

Interruptions to the ground plane under conductors carrying current can increase loop area by diverting
the return current, thus increasing loop size and facilitating ground bounce. This is especially
troublesome in a double-sided board.
The same thing applies for signal traces where the increased inductance and loop area can lead to
increased inductive coupling of unwanted interference signals. This obviously should be avoided.
Note that the dc path is more direct than the ac path. As frequency increases the path of least resistance
becomes the path of least impedance. At high frequencies return path currents tend to concentrate under
the forward path (remember that all signals are really loops) to reduce inductance and therefore
impedance.
Multilayer boards are almost always used in modern high-density layouts, and the ground plane breaks
are much less than with simple double-sided boards.

4.71
Hardware Design Techniques

Effects of Component Orientation

In this example, a two-layer PCB is constructed so that a bypass capacitor is attached at right angles or
in-line to a top-layer supply line. In the example, the ground plane is solid and uncut. Power trace
current (on the top side) flows through the capacitor, down the via, and out the ground plane.
Because ac current always takes the path of least impedance, ground return current on the lower example
rounds the corner on its way back to the source. So the current’s magnetic field and the associated loop
area change when either magnitude or frequency of the current changes, hence the changing flux. The
tendency of current to flow along the easiest path means that even a solid-sheet ground plane can have
ground bounce—irrespective of its conductivity.

4.72
Hardware Design Techniques

Shielding

4.73
Hardware Design Techniques

Reflection and Absorption Are the Two


Principal Shielding Mechanisms
Reprinted from EDN Magazine (January 20, 1994), © CAHNERS PUBLISHING COMPANY 1995, A Division of
Reed Publishing USA

INCIDENT RAY

REFLECTED RAY

TRANSMITTED
RAY
SHIELD
MATERIAL

ABSORPTIVE
REGION

Applying the concepts of shielding requires an understanding of the source of the interference, the
environment surrounding the source, and the distance between the source and point of observation (the
receptor or victim). If the circuit is operating close to the source (in the near-, or induction-field), then
the field characteristics are determined by the source. If the circuit is remotely located (in the far-, or
radiation-field), then the field characteristics are determined by the transmission medium.
A circuit operates in a near-field if its distance from the source of the interference is less than the
wavelength (λ) of the interference divided by 2π, or π/2λ. If the distance between the circuit and the
source of the interference is larger than this quantity, then the circuit operates in the far field. For
instance, the interference caused by a 1 ns pulse edge has an upper bandwidth of approximately
350xMHz. The wavelength of a 350 MHz signal is approximately 34 inches (the speed of light is
approximately 12"/ns). Dividing the wavelength by 2π yields a distance of approximately 5 inches, the
boundary between near- and far-field. If a circuit is within 5 inches of a 350 MHz interference source,
then the circuit operates in the near-field of the interference. If the distance is greater than 5 inches, the
circuit operates in the far-field of the interference.

4.74
Hardware Design Techniques

Conductivity and Permeability for


Various Shielding Materials
MATERIAL RELATIVE RELATIVE
CONDUCTIVITY PERMEABILITY

Copper 1 1

Aluminum 1 0.61

Steel 0.1 1,000

Mu-Metal 0.03 20,000

Conductivity: Ability to Conduct Electricity

Permeability: Ability to Absorb Magnetic Energy

For electric fields the primary shielding mechanism is reflection loss, and at high frequencies, the
mechanism is absorption loss. For these types of interference, high conductivity materials, such as
copper or aluminum, provide adequate shielding.
At low frequencies, both reflection and absorption loss to magnetic fields is low; thus, it is very difficult
to shield circuits from low-frequency magnetic fields. In these applications, high-permeability materials
that exhibit low reluctance provide the best protection. These low-reluctance materials provide a
magnetic shunt path that diverts the magnetic field away from the protected circuit.

4.75
Hardware Design Techniques

Thermal Design

4.76
Hardware Design Techniques

Thermal Design Basics

‹ θ = Thermal Resistance (°C/W)


TA
‹ ΔT = P × θ
‹ θJA = Junction-to-Ambient Thermal Resistance θCA
‹ θJC = Junction-to-Case Thermal Resistance
‹ θCA = Case-to-Ambient Thermal Resistance TC
‹ θJA = θJC + θCA
+ θJC
‹ TJ = TA + (P × θJA), P = Total Device Power Dissipation
‹ TJ (Max) = Data sheet parameter. Varies device to
TJ
device.

The basic concept of thermal design is to keep the junction temperature of the chip below its rated
maximum junction temperature. The maximum rated junction temperature varies from device to device,
but is generally between 125°C and 150°C.
The change in temperature is analogous to a resistance. The amount of power times the thermal
resistance equals the temperature rise. The power is analogous to a current, and the thermal resistance is
analogous to a resistance (obviously).
The thermal resistance is typically divided up into two components. The first is the thermal resistance
from the junction to the case (θJC) and the thermal resistance from the case to the ambient (θCA). The θJC
is determined primarily by the package. It is typically higher for smaller packages. The θCA can be
reduced by the addition of heatsinks. The θJA number is generally given in still air (i.e. no fans). Adding
airflow will further reduce the thermal resistance somewhat.
θJA is very dependent on PCB area, copper volume, and the dissipation of components around it. It is
not incredibly useful because the data sheet will specify this number at one or possibly two points,
which are very unlikely to be exactly like many other applications. While using θJC to predict die
temperature is not perfect, it is significantly better than using θJA. One simply measures the temperature
of the component using thermocouples or a thermal imaging device, adds that to θJC × P (estimated
power dissipation of the IC) to find the junction temperature or θJC× P + TC = TJ, where TC is the case
temperature. Note that even on a small IC, there can be “hot spots,” so it is advisable to "hunt" around
that IC using thermocouples for the “hot spot.” A thermal imaging device should clearly define that spot.
ADI has an online tool to help with this calculation:
www.analog.com/Analog_Root/static/techSupport/designTools/interactiveTools/powertemp/powertemp.
html
This tool can be accessed form the Design Center tab on the ADI website.

4.77
Hardware Design Techniques

ADP1706 1A LDO Power Dissipation


500 mm2 of PCB Copper, TA = 25°C, SOIC
SOIC (TOP VIEW)

1A
750mA
500mA

300mA

LFCSP (TOP VIEW) Exposed


Paddle 100mA

10mA

1mA

Both packages have an


exposed paddle on the
bottom which should
be soldered to the
copper ground plane

In some packages there is a lug or paddle that is meant to be soldered to the PCB. This is a means of
decreasing the thermal resistance from the die to the PCB. The lug or paddle is typically the pad to
which the chip is bonded and is in direct thermal contact with the die. The larger the copper land area
that this lug is soldered to, the better the PCB behaves as a heatsink. Increasing copper thickness and
area also increases the ability of the PCB to extract heat out of the IC.
This figure shows the thermal derating curves for the ADP1706, a 1 A LDO regulator. It is available in
two packages, both of which have an exposed paddle on the bottom which should be soldered to a
copper ground plane (1 ounce copper thickness). The data in the plot assumes a 500 mm2 copper area
and an ambient temperature of 25°C. The plot shows the junction temperature as a function of the
difference between the input and output voltage for various load currents. The maximum allowable
junction temperature for this device is 125°C.

A good reference for dealing with exposed paddle packages can be found at www.analog.com:

Gary Griffin, "A Design and Manufacturing Guide for the Lead Frame Chip Scale Package (LFCSP),"
Application Note AN-772, Analog Devices, 2006.

4.78
Hardware Design Techniques

Thermal Resistance of Popular Packages


PACKAGE θJA COMMENTS
(ºC/W)
3-lead SOT-23 300
5-lead SOT-23 190
6-lead SOT-23 165
6-lead TSOT 186 ADP1864 SW Controller
8-lead SOIC 160
16-lead QSOP 105 ADP1821 SW Controller
8-lead MSOP 75 2
ADP1715/ADP1716 LDOs (500mm copper)
8-lead LFCSP (*EP) 60 2
ADP1706/ADP1707/ADP1708 LDOs (500mm copper)
8-lead SOIC (*EP) 51 2
ADP1706/ADP1707/ADP1708 LDOs (500mm copper)
16-lead LFCSP (*EP) 20-40 ADP2105/ADP2106/ADP2107 SW Regulators
ADP1740/ADP1741 LDOs
SOT-223 65 ADP3338/ADP3339 LDOs
2
(1 in copper), θJC = 27°C/W
TO-220 35 2
(1 in copper), θJC = 3°C/W
TO-263 (D2PAK) 35 2
(1 in copper), θJC = 3°C/W
*EP = Exposed Paddle
Above values are typical, consult product data sheet for exact value

Here are some example θJA numbers for some popular IC packages, especially those commonly used for
power circuits.
It should be noted that there can be some variation in these numbers between different manufacturers
depending upon the method of measurement, PCB mounting, etc.
See also:
www.analog.com/Analog_Root/static/Packages/ThermalDataWP.pdf
This can also be accessed form the Package Thermal Characteristics tab on the heat sink design tool.
www.analog.com/Analog_Root/static/techSupport/designTools/interactiveTools/powertemp/powertemp.
html

4.79
Hardware Design Techniques

Heat Sink Basics


HEAT SINK TA
TA
TS θSA

θJA = θJC + θCS + θSA


TS
IC TS ≈ TC, θCS ≈ 0
θCS
TC θJA ≈ θJC + θSA
TJ TC
θSA ≈ θJA – θJC +

θJC
PD = DEVICE POWER DISSIPATION
TJ(MAX) = MAXIMUM JUNCTION TEMPERATURE
TJ
TA(MAX) = MAXIMUM AMBIENT TEMPERATURE
TJ(MAX ) − TA (MAX ) TJ(MAX ) − TA(MAX )
θ JA = θ SA ≈ − θ JC
PD PD

The fundamental purpose of heat sinks and airflow is to allow high power dissipation levels while
maintaining safe junction temperatures. There are many tradeoffs which can be made between airflow
and heat sink area, and this section examines some of them.
A thermal model of an IC and a heat sink is shown in this figure. The critical parameter is the junction
temperature, TJ, which must be kept below 150ºC for most ICs. The model shows the various thermal
resistances and temperatures at various parts of the system. TA is the ambient temperature, TS is the heat
sink temperature, TC is the IC case temperature, and TJ is the junction temperature. The heat sink is
usually attached to the IC in such a manner as to minimize the difference between the IC case
temperature and the heat sink temperature. This is accomplished by a variety of means, including
thermal grease, machined surface contact area, etc. In any case, the thermal resistance between the heat
sink and the IC case can usually be made less than 1ºC/W to 5°C/W.

4.80
Hardware Design Techniques

TO-220 and TO-263 (D2PAK) Packages

TO-263 G

D 10mm
S

θJA = 73°C/W
TO-220
θJC = 3°C/W
G
D

S 0.4"
10.16mm

θJA = 53°C/W
θJC = 3°C/W

This figure shows two standard packages which are capable of dissipating considerable power
Also note that the drain (D) connection for the TO-263 package is electrically connected to the slug.
This means that electric isolation from the heatsink is required for safety.

4.81
Hardware Design Techniques

Thermal Resistance of TO-263 (D2PAK) vs.


Drain Pad Area—No Airflow
70
Courtesy
AAVID Thermal Technologies, Inc.
60
θSA
(°C/W) 50

40

30

0 0.5 1.0 1.5 2.0


(0) (323) (645) (967) (1290)

DRAIN PAD PCB AREA, in2 (mm2)

The thermal resistance of a TO-263 package soldered to a PCB land area. Obviously, a larger PCB land
area will make a better heatsink. Copper is a fairly effective conductor of heat (as well as electricity).
Again, this is in still air. Airflow will help some (see next page).
This figure shows the thermal resistance of the TO-263 package as a function of PC board drain pad area
which is acting as the heat sink. Note that even with 2 square inches of pad area, the thermal resistance
is still 30ºC/W.

4.82
Hardware Design Techniques

Thermal Resistance of AAVID 573300


Surface Mount Heat Sink vs. Airflow
25

1" TO-263,
PAD AREA > 1.6in2
20 NO HEATSINK
θSA
(°C/W) 15

10
AAVID 573300
HEAT SINK
Courtesy 5
AAVID Thermal
Technologies, Inc.
0
0 100 200 300 400 500
AIRFLOW (LFPM)

The situation can be improved by the addition of a surface-mount heat sink as shown in this figure
(AAVID part number 573300). This heat sink solders to two pads on the PC board which are extensions
of the drain pad connecting area. The thermal resistance of this combination as a function of airflow is
shown in this figure.
Note that with the addition of the surface-mount heat sink, the thermal resistance of the combination is
reduced to approximately 10ºC/W with a reasonable amount of airflow (200 linear feet per minute). The
curve also shows the thermal resistance with no heat sink as a function of airflow, clearly indicating that
a heat sink definitely is effective for high power dissipation.

4.83
Hardware Design Techniques

Technical References

4.84
Hardware Design Techniques

Analog Devices' Textbook References


1. Hank Zumbahlen, Basic Linear Design, Analog Devices, 2006, ISBN: 0-915550-28-1.
Also available as Linear Circuit Design Handbook, Elsevier-Newnes, 2008, ISBN-10:
0750687037, ISBN-13: 978-0750687034. See Chapter 12.

2. Walt Kester, Analog-Digital Conversion, Analog Devices, 2004, ISBN: 0916550273.


Also available as Data Conversion Handbook, Elsevier-Newnes, 2005, ISBN:
0750678410. See Chapter 9.

3. Walt Jung, Op Amp Applications, Analog Devices, 2002, ISBN: 0-916550-26-5. Also
available as Op Amp Applications Handbook, Elsevier-Newnes, 2004, ISBN: 0-7506-
7844-5. See Chapter 7.

4.85
Hardware Design Techniques

Other Textbook References


1. Ralph Morrison, Grounding and Shielding Techniques, 4th Edition, John Wiley, Inc., 1998, ISBN:
0471245186.

2. Henry W. Ott, Noise Reduction Techniques in Electronic Systems, 2nd Edition, John Wiley, Inc.,
1988, ISBN: 0-471-85068-3.

3. Howard W. Johnson and Martin Graham, High-Speed Digital Design, PTR Prentice Hall, 1993,
ISBN: 0133957241.

4. Howard W. Johnson and Martin Graham, High Speed Signal Propagation: Advanced Black Magic,
Prentice Hall, 2003, ISBN: ISBN-10: 013084408X, ISBN-13: 978-0130844088.

5. Ralph Morrison, Solving Interference Problems in Electronics, John Wiley, 1995.

6. C. D. Motchenbacher and J. A. Connelly, Low Noise Electronic System Design, John Wiley, 1993.

7. Mark Montrose, EMC and the Printed Circuit Board, IEEE Press, 1999 (IEEE Order Number
PC5756).

8. Eric Bogatin, Signal Integrity–Simplified, Prentice Hall, 2003, ISBN-10: 0130669466, ISBN-13: 978-
0130669469.

9. Stephen H. Hall, High-Speed Digital System Design: A Handbook of Interconnect Theory and
Design Practices, Wiley-IEEE Press, 2000, ISBN-10: 0471360902, ISBN-13: 978-0471360902.

4.86
Hardware Design Techniques

Notes:

4.87

You might also like