Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
7
APRIL
Issue 7/2011
Band together Pg 56
EDN.comment: The
3-D IC and you Pg 8
www.edn.com
Signal Integrity:
Whang that ruler Pg 16
Design Ideas Pg 45
Make your electronics
supply chain green—
VOICE OF THE ENGINEER or else Pg 52
ARM
UNDERSTANDING
THE IMPACT OF
DIGITIZER NOISE
VERSUS ON OSCILLOSCOPE
MEASUREMENTS
Page 18
INTEL: INNOVATIVE
CIRCUIT DESIGNS
A SUCCESSFUL TARGET
PERFORMANCE
STRATAGEM IMPROVEMENT
FOR RISC OR AND PRODUCT
DIFFERENTIATION
GRIST FOR Page 36
Bandwidth (MHz) 70, 100, 200 50, 70, 100, 200 100, 200, 350, 500 100, 200
Oscilloscopes Max memory depth 100 kpts 2.5 kpts 4 Mpts 1 Mpt
>OLU`V\HYLYLHK`[VZWLLK\W`V\Y[LZ[PUNJVTLZLL\ZH[^^^JLU[LSSH_JVT
a successful stratagem
Understanding the Innovative circuit
for RISC or grist for
impact of digitizer designs target
CISC’s tricks?
noise on oscilloscope performance
ARM and its licensees are
pulse
Accenture
Dilbert 11
DESIGNIDEAS
45 Low-component-count logic probe works with TTL and CMOS logic
EDN® (ISSN#0012-7515) is published semimonthly, 24 times per year, by UBM Electronics, 11444 W. Olympic Blvd., Los Angeles, CA 90064-
1549; 310/445-4200; FAX 310/445-4299. Periodicals postage paid at Los Angeles, CA, and at additional mailing offices. SUBSCRIPTIONS—
Free to qualified subscribers as defined on the subscription card. Rates for nonqualified subscriptions, including all issues: US, $150 one year;
$250 two years; $300 three years. Except for special issues where price changes are indicated, single copies are available for $10 US and $15
foreign. For telephone inquiries regarding subscriptions, call 763/746-2792. E-mail: EDN@kmpsgroup.com. CHANGE OF ADDRESS—Notices
should be sent promptly to PO Box 47461, Plymouth, MN 55447. Please provide old mailing label as well as new address. Allow two months
for change. NOTICE—Every precaution is taken to ensure accuracy of content; however, the publisher cannot accept responsibility for the cor-
rectness of the information supplied or advertised or for any opinion expressed herein. POSTMASTER—Send address changes to EDN, PO
Box 47461, Plymouth, MN 55447. Canada Post: Publications Mail Agreement 40612608. Return undeliverable Canadian addresses to Pitney
Bowes Inc, PO Box 25542, London, ON N6C 6B2. Copyright 2011 by United Business Media. All rights reserved. Reproduction in whole or THE POWER MANAGEMENT LEADER
part without written permission is prohibited. Volume 56, Number 7 (Printed in USA).
APRIL 7, 2011 | EDN 7
EDN.COMMENT
O
ance. There’s a lot of work that must
has been the 3-D IC. Panels and papers cover a huge range occur before we reach that ideal,
of topics, but they come down to three questions: What is however.
a 3-D IC, is it real, and what difference does it make? The Clearly, stacking works today, if one
question of definition is surprisingly loaded. At a recent partner—typically, a giant foundry—is
panel, speakers divided the world of 3-D ICs into three cat- tightly controlling a small supply chain.
Technology partitioning is more prob-
egories. The first category covers simply stacking up independently designed
lematic. Samsung recently demonstrat-
dice and bonding them together, such as the stack of flash and DRAM dice
on the SOC (system-on-chip) die in your cell phone. In the stacking ap-
proach, all the dice are pretested standard parts, often simply wire-bonded
Discussions of 3-D
together using their normal I/O bonding pads, sometimes with a silicon in- ICs come down to
terposer to move signals around for the best wire-bonding layout. three questions:
The second category of 3-D ICs starts in the 3-D space they create by stack- What is a 3-D IC, is it
in the architectural-design phase of the ing many dice together. TSVs become real, and what differ-
system. Architects and IC designers just another element in the routing hi-
partition the system according to the erarchy. In this mindset, a cell may land ence does it make?
best technology in which to implement in one corner of the third die in the
each block. Logic blocks might go into stack—not because of its process tech- ed a wide-I/O DRAM die stacked on a
a 20-nm logic process; bulk memory, nology but simply because that lo- logic die bearing TSVs, showing that
into a DRAM process; and I/Os cation optimizes timing for the this approach is not science fiction.
and other AMS (analog/ nets in the area. You can Simon Burke, a Xilinx distinguished
mixed-signal) blocks, imagine the com- engineer, says that his company has in-
into a large-geom- plexity of a place- ternally produced an arrangement that
etry, higher-volt- ment algorithm stacks a dense FPGA-logic fabric in
age process. The that must keep one technology atop a lower-density
team would design track of the addi- AMS die. Many issues still exist; these
COMPOSITE IMAGE BY TIM BURNS: BLOCK PATTERN: BETACAM-SP/ISTOCKPHOTO.COM
one die in each tional delays and issues include diverse tool chains, lack
process, optimiz- space required for of good TSV models, absence of ther-
ing the interfaces the TSVs, the mal and mechanical 3-D modeling
among the dice for thermal- and me- tools, and lack of standards.
performance needs chanical-stress As for true 3-D, neither design nor
and power con- impacts of the cell analysis tools exist to support it, and
straints. They would on its local neigh- neither the TSVs nor the necessary
then assemble the dice borhood, and so on. die-thinning techniques are ready. KK
into a stack, using flip-chip Each of these categories Lin, Samsung’s director of foundry de-
technology, interposers, and TSVs has its own drawbacks and bene- sign enablement, hopes that his compa-
(through-silicon vias) to interconnect fits. Stacking primarily delivers great- ny will have wide I/Os available to de-
the blocks. One panelist calls this ap- er component density for space-con- signers in 2013, but true 3-D could still
proach technology partitioning. strained boards. Technology partition- lag years behind that. In short, 3-D will
In the third category, which you ing also offers greater density, but it fur- come—but slowly and in waves.EDN
could call true 3-D, designers place ther hints at cost savings and signifi-
each cell not in a plane on one die but cant performance improvements from Contact me at ron.wilson@ubm.com.
support
1-925-736-7617; Bonnie Baker,
judy.hayes@ubm.com Texas Instruments
EDITORIAL DIRECTOR Pallab Chatterjee,
they’ve
Ron Wilson, SiliconMap
1-415-947-6317; Kevin C Craig, PhD,
ron.wilson@ubm.com Marquette University
MANAGING EDITOR
Amy Norcross
LEAD ART DIRECTOR
Marco Aguilera
given us…
has been
Contributed technical articles ASSOCIATE ART DIRECTOR
1-781-869-7971; Tim Burns
amy.norcross@ubm.com
PRODUCTION
exceptional.”
MANAGING EDITOR⎯NEWS Jeff Tade,
Suzanne Deffree Production Director
Electronic Business, Distribution Brian Wu,
1-631-266-3433; Publications Production Manager
suzanne.deffree@ubm.com Jeff Polman,
Derric Treece,
Director of Engineering,
SENIOR TECHNICAL EDITOR
Brian Dipert Senior Production Artists Medical Interconnect Manufacturer
Consumer Electronics, William Baughman,
Multimedia, PCs, Mass Storage Lucia Corona,
1-916-548-1225; Ricardo Esparza,
brian.dipert@ubm.com Production Artists
TECHNICAL EDITOR EDN EUROPE
Margery Conner Graham Prophet,
Power Sources, Components, Editor, Reed Publishing
Green Engineering gprophet@reedbusiness.fr
1-805-461-8242;
MAXIMUM
EDN ASIA
margery.conner@ubm.com Wai-Chun Chen,
TECHNICAL EDITOR Group Publisher, Asia
Service
Mike Demler waichun.chen@ubm.com
EDA, IC Design and Application Kirtimaya Varma,
1-408-384-8336; Editor-in-Chief
mike.demler@ubm.com kirti.varma@ubm.com
TECHNICAL EDITOR EDN CHINA
Paul Rako
Analog, RF, PCB Design
1-408-745-1994;
William Zhang,
Publisher and Editorial Director
william.zhang@ubm.com
“ …they make things simple. Where others
find ways to complicate the process, Mill-Max
paul.rako@ubm.com Jeff Lu,
Executive Editor consistently delivers what’s needed, when
”
DESIGN IDEAS EDITOR
jeff.lu@ubm.com
Martin Rowe, Senior Technical Editor, it’s needed, with little or no fuss.
Test & Measurement World EDN JAPAN
edndesignideas@ubm.com Katsuya Watanabe,
SENIOR ASSOCIATE EDITOR
Publisher At Mill-Max we do everything we can to make your
Frances T Granville, 1-781-869-7969;
katsuya.watanabe@ubm.com
Ken Amemoto,
job easier — from our easy-to-use website to
frances.granville@ubm.com
Editor-in-Chief quick-turnaround quotes…responsive customer
ASSOCIATE EDITOR ken.amemoto@ubm.com
Jessica MacNeil, 1-781-869-7983;
EXECUTIVE OFFICERS
service and technical support…on-time delivery…
jessica.macneil@ubm.com
Paul Miller, free samples…and so much more.
CONSULTING EDITOR Chief Executive Officer,
Jim Williams,
Staff Scientist, Linear Technology
edn.editor@ubm.com
UBM Electronics
and UBM Canon (Publishing Division)
David Blaza,
“andQuality is not open to compromise,
Mill-Max components meet our
very stringent standards.”
CONTRIBUTING TECHNICAL EDITORS Senior Vice President,
Dan Strassberg, UBM Electronics
strassbergedn@att.net Karen Field,
Nicholas Cravotta, Senior Vice President, Content,
UBM Electronics
®
editor@nicholascravotta.com
Robert Cravotta, Our connectors speak for
robert.cravotta@embeddedinsights.com
themselves…so do our customers.
EDN. 33 Hayden Avenue, Lexington, MA 02421. www.edn.com. Subscription
inquiries: 1-763-746-2792; EDN@kmpsgroup.com. Address changes: Send
notice promptly to PO Box 47461, Plymouth, MN 55447. Please provide an old
mailing label as well as your new address. Allow two months for the change.
TALKBACK
AWG boasts 14-bit resolution, “I designed some
12G-sample/sec updates of the first auto-
matic toilet flush-
gilent Technologies has added a high- Express) architecture, which targets use in ers in 1983 so
A resolution, wide-bandwidth, 8- or
12G-sample/sec modular instrument
to its AWG (arbitrary-waveform-generator)
high-performance instrumentation, reduces
system size, weight, and footprint.
Agilent’s Measurement Research Lab
had to deal with
sifting microwatts
portfolio. The M8190A enables radar-, satel- designed a proprietary DAC for the genera- out of a sea of
lite-, and electronic-warfare-device designers tor using an advanced silicon-germanium garbage.”
to make reliable, repeatable measurements bipolar/CMOS process. The DAC operates at —Electronics designer Robert
and create highly realistic signal scenarios to 8G samples/sec with 14-bit resolution and at Capper, in EDN’s Talkback sec-
test their products. Precision arbitrary-wave- 12G samples/sec with 12-bit resolution. At 8G tion, at http://bit.ly/himngN.
Add your comments.
form generation is necessary for realistic test- samples/sec, the DAC delivers SFDR (spuri-
ing of systems for detecting low-flying aircraft ous-free dynamic range) of as much as 80 dBc
and for high-data-rate communications in sat- (decibels referenced to the carrier). This tech-
ellite-communications systems. With resolu- nology eliminates the trade-off between high
tion as high as 14 bits, the M8190A makes resolution and wide bandwidth, so measure-
it easy for designers to distinguish between ments are more reliable and repeatable and
signals and distortion in their test scenarios you are less likely to misinterpret glitches in
and to more rigorously stress their devices. the waveforms as analog output. The modular
The instrument’s optional 2G-sample memory generator, whose US entry price is $79,000,
lets you create longer and more realistic test works in either two- or five-slot AXIe chassis.
scenarios. —by Dan Strassberg
“The M8190 allows engineers to approach ▷Agilent Technologies,
reality when they create test scenarios,” says www.agilent.com/fi
g nd/M8190.
Jürgen Beck, general manager of Agilent’s
digital- and photonic-test business. “Because
the generator simultaneously offers greater
fidelity, higher resolution, and wider band-
width and produces multilevel signals with
programmable ISI [intersymbol interference]
and jitter at frequencies to 3 Gbps, custom-
ers can create signal scenarios that push their
designs to the limit and bring new insights to
their analysis.”
The M8190A simultaneously offers 14 bits of
resolution, as much as 5 GHz of analog band-
width, and the ability to build realistic scenar- Fitting into this two-slot enclosure, which connects to a separate laptop, the
ios with 2G samples per channel of waveform M8190A AWG provides a combination of high resolution, wide bandwidth, deep
memory. Compact modular AXIe (Advanced memory, and flexible waveform-segment sequencing and programming. It also
Telecommunications Computing Architec- fits into a five-slot AXIe enclosure, in which a plug-in computer module makes it
ture Extensions for Instrumentation and Test a stand-alone unit.
R Micron (www.micron.
com) unveiled its 6-Gbps,
SATA (serial-advanced-tech-
REPLACEMENTS
In the United States, it’s
easy to see the world
nology-attachment)-supportive of LED lighting through
C300 solid-state drives and six Edison-bulb-illuminated
months after the company fol- glasses, but these bulbs
lowed them with the enterprise- are not the dominant
targeted P300 variants, Intel global light sources.
has finally rolled out a 6-Gbps Europe, for example, has
SATA-cognizant solid-state- more than 1 billion 35 to
drive family. Whereas Micron’s 50W MR-16 halogen lights,
latest-generation C400 drives about 2 inches long and
Intel’s 6-Gbps SATA-cognizant 510 series of solid-state
employ NAND devices built on selling for about $12 each.
drives touts higher sequential-read and -write speeds than
a 25-nm process that the com- With the replacement
its predecessor.
pany jointly developed with market for these LEDs in
IM Flash Technologies (www. products, such as the X25-M Mbytes/sec, respectively, pre- mind, Cree recently intro-
series, also employed SATA- sumably due to the fact that the duced the MT-G LED array.
The transi- system-interface IP (intellectual controller can simultaneously It combines the company’s
tion to 6- property) from Marvell. Intel’s access fewer storage-array EasyWhite color-mixing
expertise came into play at the components and the blocks technology, a thermal
Gbps SATA has other end of the SATA-logic within those components. And resistance of 1.5°C/W,
a notable effect block, extending to the embed- the transition from 3- to 6-Gbps and efficacy as high as
on sequential- ded flash-memory array.
A notable downside of Intel’s
SATA also has a notable effect
on sequential-access perform-
92 lm/W (560 lm at 6W) at
85°C (3000K) in a 9.1×9.1-
access earlier-generation, 3-Gbps ance: When the 250-Gbyte mm footprint.
performance. SATA-based solid-state drives solid-state drive connects to These devices let you
was their comparatively slow a 3-Gbps SATA bus, the drive design an MR-16 light
imftech.com), Intel has cho- sequential-write performance specifies sequential-read and with one LED, packing the
sen the more conservative, versus competitors’ products. -write speeds of 265 and 240 LED, a secondary optic,
although perhaps less cost- Judging from the 500-Mbyte/ Mbytes/sec, respectively, the power-management
effective, path of sticking with sec sequential-read and 315- whereas its 120-Gbyte sib- circuit, and a heat sink
34-nm-fabricated flash-mem- Mbyte/sec sequential-write ling clocks in at 265 and 200 into the MR-16 form fac-
ory components. speed claims for the higher- Mbytes/sec, respectively. Now tor. Cree has published a
Intel’s 510 series leverages capacity, 250-Gbyte 510 series in production, the 250- and reference design, including
the same Marvell (www.marvell. variant, the company has heard 120-Gbyte versions of the 510 power-control circuits.
com) controller that Micron’s and responded to the grum- series sell for $584 and $284 —by Margery Conner
products use. Rumor has it bling. Sequential-read and (1000), respectively. ▶Cree, www.cree.com.
that Intel’s internally developed -write specifications for the 120- —by Brian Dipert
controllers for earlier-generation Gbyte product are 400 and 210 ▷Intel Corp, www.intel.com.
04.07.11
Tool for custom-IC design allows verification in real time
Mentor Graphics’ new Calibre RealTime platform al- Calibre RealTime executes direct calls to Calibre
lows designers to execute physical verification in analysis engines running foundry-qualified rule decks.
real time during IC-layout creation. The first release User-defined custom filters allow designers to limit
provides instantaneous DRC (design-rule checking) in which checks to run, depending on design require-
the SpringSoft (www.springsoft.com) Laker Version OA ments and organizational processes, without modifying
the foundry-qualified rule deck. When you integrate
Calibre RealTime into a custom-IC-design and -layout
system, the tool’s engines perform incremental check-
ing near the shapes you are editing, providing imme-
diate feedback on design-rule violations. By running
with the same rule set that you use for design sign-off,
Calibre RealTime complements many layout editors’
built-in checkers.
A built-in error-review tool bar in the layout-design
environment enhances ease of use. In-memory check-
ing optimizes performance. By combining Calibre
RealTime with batch Calibre and Calibre RVE (results-
viewing environment), layout designers can minimize
You can use Calibre RealTime to detect and define sign-off the need for full-chip verification runs, shortening the
DRC errors during manual layout editing, automatic routing, production schedule. Joseph Sawicki, vice president
or generation of automated P cells.
and general manager of the design-to-silicon division
(Open Access) 2010.8 custom-IC-design and -layout of Mentor Graphics, says that the OA runtime model
tools, using the same Calibre decks as for a batch sign- will enable integration with most custom design envi-
off flow. A version of Calibre RealTime for the Mentor ronments, including Cadence Virtuoso.
IC Station Version 10 custom-design environment will —by Mike Demler
become available in June. ▶Mentor Graphics Corp, www.mentor.com.
C
1-D-wave-propagation medium. If the
the edge of the desk by about 8 inches. Now, flick the end trace is long enough and if it lacks any
of the ruler (Figure 1). It resonates, doesn’t it? You can good energy-absorbing devices, it will
easily change the resonant frequency. Tape a few quarters resonate, distorting your signals.
When I say a trace has a simple, lin-
to the end of the ruler and observe that the resonant fre-
ear topology, I mean that it is a point-
quency decreases. Shorten the length of overhang and hear to-point connection or, at most, a lin-
it increase. If you push the resonant frequency high enough, it becomes ear-bus structure with multiple trans-
difficult to stimulate the resonance with the soft end of your fingertip. ceivers arrayed along a single trace.
Overcome that difficulty by depressing the end of the ruler with your fin- More complex structures, such as H
ger in a way that lets the ruler slip off the hard edge of your fingernail. distributions, star clusters, or random
hairball nets, support multiple modes
At the risk of annoying your co- of oscillation and may, like a drum-
workers, whang the ruler over and over head, require terminations in multiple
while you adjust the amount of over- locations.
hang until you get a nice, musical res- When I say “long enough,” I mean
onance at about 100 Hz. Now, dis- that the end-to-end trace delay is a sig-
connect the clamp and hold the same nificant fraction of the signal rise or
length of ruler in your hand. Flick the fall time. A delay as long as one-sixth
end. Try to create the same resonant ef- the rise or fall time is significant, es-
fect. I bet you can’t do it. pecially if the trace has a particular-
When you clamp the ruler to the ly low-impedance driver or a large ca-
desktop, the clamp creates a low, me- pacitive load. I simulate all such trac-
chanical impedance at one end of the es. Keep in mind that even short traces
ruler. The other end of the ruler re- Figure 1 Whanging a clamp-attached resonate; the resonant frequency is just
mains free to move—the only limit ruler causes it to oscillate for many so high that you may not observe its ef-
being air resistance. Your stimulation cycles before settling down. fect using logic with a particular rise
creates a mechanical wave that bounces and fall time. Applying logic with fast-
back and forth between these two end- er edges to the same trace might make
points, neither of which absorbs much and back again. Mechanical engineers it ring, just as flicking the ruler with
of the mechanical energy. It therefore call that phenomenon 1-D wave propa- your fingernail stimulates resonance at
takes many back-and-forth cycles for gation. For any system like this one, an a small length.
the ruler to settle down. You have cre- absorbing device at either end can to- Applying a capacitive load to a PCB
ated a highly resonant system. tally damp the oscillations. Mechan- trace has much the same effect as a load
When you hold the ruler in your ical engineers use hydraulic shock ab- of quarters on the end of the ruler: It
hand, the mechanical impedance of sorbers, friction, air resistance, and rub- lowers the resonant frequency of the
your hand lies close to the natural char- ber to absorb energy and create damp- structure, making it more likely that
acteristic impedance of the ruler. Even ing. Electrical circuits use resistive you will notice its effects.EDN
if it is not a perfect match, your hand terminations.
absorbs a significant portion of the en- A more complex system, such as a Howard Johnson, PhD, of Signal Con-
ergy in each cycle. The result is that child’s Indian drum, supports wave mo- sulting, frequently conducts technical
the ruler cannot resonate. tion in two dimensions. Waves on the workshops for digital engineers at Oxford
The ruler supports transverse me- surface of the drumhead spread and re- University and other sites worldwide. Visit
chanical waves in just two directions: flect in many highly varied and com- his Web site at www.sigcon.com, or e-mail
from one end of the ruler to the other plex patterns. An absorbing device him at howie03@sigcon.com.
NE W
O
ING OR BUYING A DIGITIZ- measurements is the presence of vertical noise,
ING SYSTEM, YOU NEED which can decrease the accuracy of signal meas-
SOME MEANS OF DETERMIN- urement and lead to such problems as inaccu-
ING REAL-LIFE PERFORM- rate measurements as frequencies change. You
ANCE. HOW CLOSELY DOES
can use ENOB (effective-number-of-bits) test-
THE OUTPUT OF ANY ADC,
WAVEFORM DIGITIZER, OR
ing to more accurately evaluate the perform-
DSO FOLLOW AN ANALOG ance of digitizing systems, including oscilloscopes. The ENOB
INPUT SIGNAL? ENOB TEST- figure summarizes the noise and frequency response of a sys-
ING PROVIDES A MEANS OF tem. Resolution typically degrades significantly as frequency in-
ESTABLISHING A FIGURE OF creases, so ENOB versus frequency is a useful specification. Un-
MERIT FOR DYNAMIC DIGI- fortunately, when an ENOB specification is provided, it is of-
TIZING PERFORMANCE. ten at just one or two points rather than across all frequencies.
5.5
EFFECTIVE
NUMBER 5
OF BITS
4.5
4
Figure 7 A test applies a 6.5-GHz sine
wave to a DPO with 13-GHz bandwidth 3.5
and 400-mV full-scale amplitude. The 0 2 4 6 8 10 12 14
DPO also has infinite display persistence FREQUENCY (GHz)
to show variations across all acquisitions. Figure 8 This result from Figure 7’s test corresponds to approximately 5.9 ENOBs at
6.5 GHz.
offer a way to restore that third dimen-
sion by grading the signal employing com) DPO/DSA70000B oscilloscope (Figure 7). This result corresponds to
the frequency of the hits (Figure 6). with a 13-GHz bandwidth and a 400- approximately 5.9 ENOBs at 6.5 GHz
mV full-scale voltage. It also has infi- (Figure 8). Comparative testing shows
REAL-WORLD SIGNAL NOISE nite display persistence so that you can other oscilloscopes with more than 35-
ENOB performance indicates noise see variations across all acquisitions. mV trace thickness at the peak and ap-
that has an effect on both amplitude With no averaging, the test run includ- proximately 4.5 ENOBs using identical
and timing measurements. To illus- ed approximately 10,000 acquisitions. test setups.
trate the effects of noise on ampli- The result is approximately 15.9 mV of
tude, a test applied a 6.5-GHz sine trace thickness at the peak, represent- ENOB EFFECTS
wave to a Tektronix (www.tektronix. ing 3% of full-scale on this oscilloscope ENOB effects can also be seen on
The S-Series
Low Phase Noise
Aeroflex introduces the signal generator of the
Fast Switching Speed future, and makes compromise a thing of the past.
RM, along with its core licensees, and Intel, Since ARM unveiled the Version 7
A
along with its x86 CPU competitors, have re- instruction set, the company has sub-
cently taken action to put to rest any remain- divided its product line into three seg-
ments: the highly integrated Cortex-A
ing doubt that both camps were on a collision application processors for mobile devic-
course—ARM touting its RISC (reduced-in- es, cost-sensitive Cortex-M processors
struction-set-computer)-based technology and for traditional microcontroller applica-
Intel backing the CISC (complex-instruction- tions, and high-performance Cortex-R
set-computer) approach. When Intel three years ago formally in- processors for deeply embedded real-
troduced the first-generation Atom processor family, the company time applications. Cellular handsets,
multimedia record-and-playback devic-
made it clear that it was aiming not just at low-end desktop and es, and other portable electronics sys-
notebook PCs but also at the handheld systems in which ARM tems incur substantial product volume
had historically dominated. In response, ARM more recently un- shipments. That fact, along with their
veiled the Cortex-A15 core, whose application targets extend up direct competition with Intel-architec-
TICKET: ALENATE/ISTOCKPHOTO.COM
to the server segment in which Intel and AMD (Advanced Mi- ture processors, explains why this article
cro Devices) have long reigned supreme. And at the January 2011 focuses on ARM Cortex-A CPUs (see
sidebar “Intel’s potential multiphase re-
CES (Consumer Electronics Show), Microsoft revealed its will- sponse”). For more, see EDN’s coverage
ingness to put a nail in the coffin of the Wintel alliance by broad- of Cortex-M, Cortex-R, and alterna-
ening upcoming Windows 8’s instruction-set compatibility to en- tive-market Cortex-A products, such as
compass both ARM and x86. Ambarella’s iOne (Reference 1).
3LFNHULQJ
5HHG5HOD\V
(c)
1HHG,QVWUXPHQWDWLRQ
(a) (b) *UDGH5HHG5HOD\V"
/RRN1R)XUWKHU
3LFNHULQJ(OHFWURQLFVLV
D0DQXIDFWXUHURIKLJK
TXDOLW\5HHG5HOD\VIRU
y ,QVWUXPHQWDWLRQ
+LJKHVW*UDGH5HHG6ZLWFKHV
y $XWRPDWLF7HVW
(TXLSPHQW$7(
y +LJKYROWDJHVZLWFKLQJ
(e) y /RZWKHUPDO(0)
(d) (f)
y 'LUHFWGULYHIURP&026
Figure 1 Nvidia’s Tegra 2 SOC is seemingly the reference y 5)VZLWFKLQJDQGRWKHU
CPU for Google’s Android 3.0 Honeycomb operating system VSHFLDOLVWDSSOLFDWLRQV
in Motorola’s Xoom tablet (a), but the integrated processor
also finds use in modern high-end smartphones, such as 6RIW&HQWHU
Motorola’s Atrix 4G (b), whose mating dock, display, and key-
0X0HWDO0DJQHWLF6FUHHQLQJ
board enable it to approximate a full PC experience (c). LG’s
Optimus 2X might use Tegra 2, but the newer Optimus 3D
went with TI’s OMAP 4 instead, due to its higher-resolution
video-encoding support (d). RIM’s BlackBerry PlayBook
(g)
2QO\3LFNHULQJKDV
tablet is another OMAP 4 design win (e), whereas Samsung’s
10.1-in. Galaxy Tab gave Tegra 2 the nod. The company’s Galaxy S II smartphone baf- 6RIW&HQWHU7HFKQRORJ\
flingly splits processor loyalties between Nvidia and Samsung’s own semiconductor
(QFDSVXODWLRQ 6RIWLQQHU
group (f). HP’s TouchPad is a Qualcomm dual-core Snapdragon advocate (g). ,QWHUQDO 6KHOO HQFDSVXODWLRQ
PXPHWDO
PDJQHWLF PDWHULDO
VFUHHQ WRSURWHFW
,QVWUXPHQWDWLRQ
and short-lived Kin cellular handset. ruary also harness its capabilities, al- *UDGH5HHG
6RIW&HQWHU
The company bypassed both the Cor- though its reported inability to decode 6ZLWFK
tex-A5 and the Cortex-A8 generations high-profile 1080p H.264-encoded vid-
to come up with its next-generation eo led to its 11th-hour replacement in 5HHG
6ZLWFK
Tegra 2 products, which it formally the Boxee set-top-box design by Intel’s
unveiled at the January 2010 CES. A Atom-based CE4100 CPU. Similarly, 'LRGH
6HOI
2QOLQH&DWDORJRU'RZQORDG
year ago, Nvidia had a somewhat-tar- LG chose a Texas Instruments OMAP VXSSRUWLQJ
nished industry reputation because the (Open Multimedia Applications Plat- FRLOWR +DUGRXWHU
PD[LPL]H HQFDSVXODWLRQ
raft of Tegra 2-based tablet computers form) 4-based CPU for its Optimus 3-D PDJQHWLF PDWHULDO
and other devices that company Chief handset, although Tegra 2 had received GULYH
Executive Officer Jen-Hsun Huang the nod in the Optimus 2X phone LG 3LFNHULQJ5HHG5HOD\VDUH
claimed at CES would shortly material- had introduced less than two months HQFDSVXODWHGXVLQJDVRIW
ize hadn’t done so. earlier. LG chose the OMAP 4 because LQQHUPDWHULDOWRSURWHFWWKH
What a difference a year makes. Nvid- TI’s SOC could encode 3-D content in UHHGVZLWFKFDSVXOH7KHYHU\
ia had achieved first-to-market status real time in 1080p resolution, whereas KDUGFRPSRXQGVXVHGE\PRVW
RWKHUPDQXIDFWXUHUVFDQFDXVH
in the dual-core Cortex-A9 generation Nvidia’s CPU had only 720p capabili-
VWUHVVHVWKDWFDQSRWHQWLDOO\
by several quarters versus competitors ties (references 3 and 4). GDPDJHWKHUHHGVZLWFKDQG
such as Texas Instruments. In doing so, Tegra 2 comes in T20 and AP20 vari- GHJUDGHFRQWDFWUHVLVWDQFH
it gained Google’s nod in the reference ants, encompassing the Tegra 230 and VWDELOLW\DQGOLIHH[SHFWDWLRQ*R
design for tablet-targeted Android Ver- Tegra 250, respectively. Both CPUs run WRZZZSLFNHULQJUHOD\FRPWR
sion 3 Honeycomb (Figure 1). Several at 1 GHz, but the GPU (graphics-pro- ¿QGRXWPRUH
high-end smartphones that debuted at cessing unit) in T20 uses a 333-MHz
the 2011 iterations of CES and MWC clock and DDR2 system memory, befit-
(Mobile World Congress) in mid-Feb- ting the larger screens and batteries in
SLFNHULQJUHOD\FRP
tab
tablets. Its handset-focused AP20 peer, speeds up to 400 MHz. It’s now unclear, rely on a unified 1-Mbyte pool of L2
in contrast, operates the GPU at 300 however, whether they’ll ever see the cache memory, which twice as many
MH
MHz and interfaces to the more-power- light of day because the quad-core fol- CPU cores as before share, as well as
stingy LP (low-power) DDR2 SDRAM. low-on AP30 and T30 SOCs, or Tegra a 32-bit system-memory interface.
Strictly speaking, it’s a triple-ARM 3, code-named Kal-El, have also ap- The Nvidia-designed GPU is not on-
configuration because it also contains peared, courtesy of an aggressive de- ly faster in Kal-El than in Tegra 2 but
an ARM7 core for overall SOC man- velopment cycle. Nvidia obtained first also ups the core count from eight to
agement. You can suspend both Cortex- Kal-El samples from its foundry partner 12. Both Tegra 2 and Kal-El employ a
A9 cores, albeit not individually, when, just 12 days before MWC, during which 40-nm manufacturing process, thereby
for example, only the audio, imaging, it was showing robust graphics and vid- explaining the die-size boost from 49
graphics, and other dedicated process- eo demos running at 2560×1600-pixel mm2 on Tegra 2 to roughly 80 mm2 on
ing resources are in use (Figure 2). resolutions, twice the per-frame pixel Kal-El. Nvidia’s marketers believe that
Nvidia leaked documentation in count of conventional 1080p video. Kal-El will deliver roughly five times
late January suggesting that it planned The target per-core Cortex-A9 clock the aggregate speed of Tegra 2—twice
to announce AP25 and T25, or Tegra speed for Kal-El is 1.5 GHz. Unlike the CPU performance and three times
2 3-D, during the first quarter of this Tegra 2, Kal-El embeds a Neon SIMD the graphics performance—and con-
year. Both variants upped the ARM vector FPU for each CPU core. Like sume no more—and, in some cases,
core clock speed to 1.2 GHz, with GPU Tegra 2, however, Kal-El continues to notably less—power on a workload-
at least the 15-nm process Instruction-set compat- sents an aspiration to con- cessors: Continued innova-
node, if not further. ibility is seemingly decreas- solidate on one operating- tion is a welcome contra-
Intel originally manu- ing in importance over time system code base, thereby diction,” EDN, Feb 4, 2010,
factured Atom on a 45-nm with consumers, due to obsoleting Windows pg 19, http://bit.ly/ik9EMK.
lithography in CPU form; the consolidating number CE-based products, ARM B Dipert, Brian, “Windows
the companion chip set of file formats that require suppliers could find them- on ARM: for Intel, probably
used a 90-nm process. It support, the increasing selves facing formidable no cause for alarm,” EDN,
has now progressed down number of OS-agnostic x86 competition in the near March 3, 2011, pg 10,
to the 32-nm process level applications that compre- future in their traditional http://bit.ly/eL4XSO.
LOW-POWER
AUDIO BACK END L3
USB-HS OTG
ARM VERSION 7 W/TRANSCEIVER IMAGING EXTENSIONS
MICRO- 466-MHz
PROCESSOR MINI64 WITH IVA-HD 200-MHz DSS THREE-PORT IMAGE- SIMCOP
2-D/3-D
WITH 32/32-kBYTE 32-kBYTE L1 1080P WITH SGX540 DDR DSI/HDMI USB-HS SDMA SIGNAL UNIPRO HSI
L1,1M L2, AND 256-kBYTE C2C DISPLAY HOST PROCESSOR
64-kBYTE RAM, 128-kBYTE L2 SL2
128-kBYTE ROM 200-MHz
IVA3 SDRC DUCATI
HIGH-BANDWIDTH INTERCONNECT
WITH MULTICHANNEL MEMORY SUPPORT
DMM C2C
Figure 3 Comparatively late to the Cortex-A9 party, Texas Instruments is determined to play catch-up with OMAP 4. The OMAP 4430
employs a PowerVR SGX 540 GPU running at approximately 300 MHz.
speeds as high as 400 MHz, and each tex-A9 cores at 1.5 GHz and delivers a integrated DLP (digital-light-projector)
CPU core includes Neon SIMD vec- 25% increase in overall graphics perform- features, as Me-D experiences that ratio-
tor floating-point capabilities in the ance, according to TI. Other enhance- nalized OMAP 4’s performance poten-
form of ARM’s MPE (media-processing ments include 60-frame/sec, 1080p vid- tial. The company also showcased Epos
engine). eo-decoding capabilities; support for as technology, which allows users to take
In December, TI also unveiled the many as two 12M-pixel camera sensors; notes and draw using a pen or a stylus.
more advanced OMAP 4440, which the an HDMI (high-definition-multimedia- The writing tool comes with an Epos-
company fabricates, like its 4430 sibling, interface) 1.4 video output; and automat- patented ultrasonic transmitter, and the
on a 45-nm process. Available for sam- ic stereoscopic 3-D displays. At MWC, OMAP processor selects the transmitted
pling now, with production slated for the TI highlighted these capabilities, along signals using three microphones to accu-
second half of this year, it runs the Cor- with gesture-based user interfaces and rately determine the location of the pen
.19"ht.
ally beneficial, as PowerVR-graphics products. Addressing that concern,
technology found its way into most the company rolled out its Series 6 low-
of ARM’s SOC (system-on-chip)- technology, code-named Rogue, at profile
based system designs. February 2011’s MWC (Mobile World
Success, however, inevitably Congress). Rogue promises a 20- to
attracts competitors. Ironically, 100-fold performance boost over
at least some of Imagination today’s quad-core graphics engines • Audio Transformers
Technologies’ competition nowa- and counts ST-Ericsson among its
days comes from ARM itself, which initial licensees. • Pulse Transformers
in mid-2006 acquired another tile- From a longer-term perspective, • DC-DC Converter
based-rendering IP provider, Falanx Imagination Technologies in mid-
Microsystems. After rebranding December announced its intention Transformers
Falanx’s products as Mali, the ARM- to acquire Caustic Graphics, a ray-
sourced graphics technology has to tracing technology pioneer. Even • MultiPlex Data Bus
date found its most visible backer if Imagination Technologies’ ARM Transformers
in Samsung, which employs it in the market share diminishes over time,
Exynos dual-core Cortex-A9 SOC. King-Smith is confident that alter- • Power &
Another PowerVR competi- native-architecture SOC designs will
tor, Qualcomm, not only does its more than pick up the slack. The EMI Inductors
own ARM CPU designs by virtue company is a key MIPS partner, for ly
e d ia t e
g im m
of its architecture license but example, and supplies the graphics- ll C a t a lo s . c o m
tronic
fu
also in 2009 purchased the for- accelerator cores that Intel’s Atom See P ic o ’s lec
picoe
mer ATI Technologies’ Xilleon chip sets use. w w w.
handheld-graphics group, which
it now rebrands as Adreno. Fellow REFERENCES or send direct for free PICO Catalog
ARM-architecture licensee Marvell A Dipert, Brian, “Getting glitzy with Call Toll Free 800 431-1064
Fax 914-738-8225
isn’t currently in Imagination graphics for embedded systems,”
E Mail: info@picoelectronics.com
Technologies’ camp, either; instead, EDN, April 1, 1999, pg 79, http://bit.
it’s leveraging the graphics exper- ly/h6OrkH.
tise of Vivante. And it’s probably no
surprise to learn that Nvidia is also
B Dipert, Brian, “Balancing in three
pliant Sheeva PJ4 CPU-core technol- ing servers. Curiously, though, at the mus Tab: Not all HD 3D video is creat-
ogy. Other differentiating features in- January 2010 CES, the company had ed equal,” SlashGear, Feb 14, 2011,
clude the number of per-SOC cores, claimed that its first quad-core ARM http://bit.ly/gEPmti.
along with their clock speeds; cache device would serve the mass consumer 4 Dipert, Brian, “Coming soon: 3-D
sizes and the number of cache levels; market and high-volume gaming appli- TV,” EDN, April 8, 2010, pg 24, http://
system-memory-bus width and speed; cations (Reference 7). bit.ly/dbBAnv.
the degree of on-chip graphics- and Speaking of the consumer market, 5 “TI’s OMAP platform spurs Me-D
video-processing horsepower; and the Marvell at MWC announced the 40- experiences: Welcome to the next
variety and number of other integrated nm-based PXA978 World Phone com- mobile dimension,” Texas Instruments,
peripherals. munications processor, running at 1.2 Feb 14, 2011, http://bit.ly/eNv3vg.
To date, Marvell has had limited GHz and including a cellular-modem 6 Galaxy S2 Specification, Samsung,
success in the computing and commu- supporting 3G UMTS (Universal Mo- http://bit.ly/ek8IKb.
nications applications that this article bile Telecommunications System) and 7 “Marvell Announces Another Break-
discusses, with the exception of being China’s TD-SCDMA (time-division- through in Chip Technology: World’s
a key supplier to RIM. The company synchronous-code-division/multiple First Quadruple Core Processor for
keeps trying, however. Last Septem- access), with HSPA (high-speed-pack- ARM Instruction Set,” Marvell, Jan 6,
ber, for example, Marvell unveiled the et-access) support.EDN 2010, http://bit.ly/eZTxNk.
Armada 628, containing three She-
eva PJ4 Cortex-A9-class CPU cores REFERENCES
in a nod to their two-issue, limited in- 1 Dipert, Brian, “Ambarella’s iOne: You can reach
struction-reordering capabilities. Two augmenting an image-processing foun- Senior Technical Editor
of the cores run at 1.5 GHz, the third dation with ever-increasing integra- Brian Dipert
clocks in at 624 MHz, and the on-chip tion,” EDN, Dec 22, 2010, http://bit.ly/ at 1-916-548-1225,
graphics can process 200 million tri- hDf8sM. brian.dipert@ubm.com,
angles/sec. Two months later, Marvell 2 “Selecting Application Processors for and www.bdipert.com.
ADVANCED4//,"/8?%$.PDF 0-
unveiled a quad-core, 1.6-GHz Sheeva Mobile Multimedia,” Berkeley Design
P
nation. It is a deliberate expression of creativity Product innovation occurs when someone uses an inven-
that follows a scientific approach and a strate- tion or a new idea to change how a current product works.
gy that adds value and closes performance gaps. Innovation can involve gradual, fundamental, or step-func-
Relying on conventional techniques to reduce tion product changes. Innovation is not about inventing
costs may fail to close performance gaps with a something new, but it should be a creative process. Innova-
competitor because the competition may have discovered a tion involves finding new and improved ways of doing things
new or an improved way of doing something. for commercialization. Following an established method for
Consumers are in an endless pursuit to do more with less innovation is significantly more reliable than simply employ-
and to get more for less. Products that can differentiate ing serendipity.
themselves from others can create competitive advantages. Many versions of innovation exist, but most include a
Innovative products differentiate themselves from the com- common set of steps (Figure 1). The organization must first
petition by providing better ways to deliver more value to validate the product need and the business case with a well-
the customer. Mature organizations know how to apply inno- defined, investigated, documented, and communicated con-
vation to all segments of product architectures—from small cept. The organization then uses creative procedures to gen-
components and packaging to the supply-chain process. A erate a plethora of ideas, which it evaluates, ranks, and se-
holistic approach to product innovation provides the best lects for further development. Selected ideas are developed
opportunity to close performance gaps, differentiate prod- into concepts and presented for further evaluation. The or-
ucts, and reduce costs. ganization also evaluates concepts, ranking and selecting
Electronic design is no stranger to innovation. The need among them for further development. Selected concepts are
to apply innovation to IC design has swelled with the de- the basis of prototypes for additional evaluation. Finally, the
ployment of ASICs. When economic advantage exists, organization determines which one of the prototypes satisfies
many electronic designs now use ASICs. This trend means the original need and therefore justifies commercialization.
that a larger proportion of product innovation involving According to the Henderson-Clark model, innovation is
electronic design must occur at the semiconductor level. separable into two dimensions (Reference 1). Modular inno-
ASIC designs are one more level away from the true voice vation requires new knowledge involving one or more com-
of the customer. This fact presents challenges to product de- ponents of a design. Architectural innovation involves new
signers because they are not typically directly involved in methods in deploying the linkage among all the components
ASIC design. of a design. Combining extents of the modular and architec-
The trend toward more ASIC usage also presents an op- tural dimensions can result in various types of innovation.
portunity for product innovation. Developers can exploit Breakthrough innovation occurs only when designers revo-
ASIC design through the selection and refinement of sub- lutionize both the modular and the architectural dimensions.
circuits and with their adaptations to new situations. Innova- For example, the advent of television remote control was a
tive circuit designs can close performance gaps and differen- radical innovation because both the modular innovation—
tiate products. Product designers and ASIC designers must the infrared devices—and the architectural innovation—the
work together to optimize their success with product innova- remote-control electronics—were revolutionary.
tion in alignment with customer priorities. Innovation may occur at multiple levels of product archi-
Process variation adversely affects every form of produc- Figure 2 Product-architecture segmentation breaks down into
tion, including that of ASICs. ICs exhibit relatively small multiple layers.
process variation at any position across a die, meaning that
changing application requirements or modifying MOS pro- nique represents the architectural innovation of a linkage for
cesses. However, these conventional approaches are some- the sensor. The subcircuit becomes a component that enables
times impractical. Automatic on-chip compensation tech- other MOS subsystems to self-adjust for process variation.
niques for process variation could be the next breakthrough To reveal the exact on-chip threshold voltage of the P-chan-
innovation involving electronic design. Although alterna- nel transistor, this innovation first requires a means of prevent-
tives to this approach exist, product innovation compels de- ing the body effect from occurring. The body effect occurs
signers to explore all such ideas. when the bulk-to-source voltage of a P-channel transistor is
To achieve breakthrough innovation for automatic com- greater than 0V. The body effect causes the threshold voltage
pensation, both the modular and the architectural dimensions to vary as a function of the bulk-to-source voltage. The source
of innovation must be revolutionary. First, you must develop channel of a P-channel transistor in its own N well, or body,
a library of sensors, involving modular innovation that can can connect directly to its N-well terminal, thereby prevent-
accurately measure a variety of application and on-chip pro- ing the body effect. As a result, the threshold voltage of two
cess parameters. These sensors represent new “components.” matched P-channel transistors at the same position of a die re-
You must then use architectural innovation to develop new mains equal regardless of how they are biased in the circuit.
circuit-design techniques to use this new information for the The threshold-voltage sensor performs by means of the
other value-added subsystems. These approaches represent a high-gain op amp, which uses negative feedback through a
new way of configuring subsystems to be self-adjusting and source follower to ensure that the voltage to the negative in-
“smart” to compensate for unfavorable operating conditions. put of the op amp remains exactly equal to one-half of the
power supply (VDD/2) in steady state. The two matched resis-
ON-CHIP MOSFET THRESHOLD VOLTAGE tors at the positive input of the op amp establish the reference
A subcircuit for the on-chip sensing of process variation voltage. Thus, the P-channel transistors, in the absence of the
is one example of modular innovation (Figure 3). This sub- body effect, have an operating point that must satisfy the fol-
circuit captures the exact value of the on-chip P-channel lowing equation: (1×W/L)(VSG1−VT0)2=(4×W/L)(VSG2−VT0)2
MOSFET threshold voltage with zero body effects, VT0. This =(4×W/L)(VSG3−VT0)2, where W/L is the transistor’s width-to-
technique presents the threshold voltage as an exact voltage length, or aspect, ratio and VT0 represents the on-chip P-chan-
quantity, a bias current that a resistor value es-
tablishes, or an exact rational-number multiple VDD
of threshold voltage. Although this version ap-
plies to N-well CMOS technologies, you can (W/L) (W/L)
Q5
also apply it to P-well, dual-tub, and bipolar- 4 (W/L)
R4
CMOS technologies.
The threshold-voltage sensor contains Q3
Q6
uniquely biased P-channel transistors and a
R
feedback loop with high gain for control. You 4 (W/L) 3
VIN
VIN
10μF VIN
2.2Ω 7V TO 14V
100k VIN 16V + +
X5R 180μF 180μF
0.1μF 16V 16V
DIFFOUT MODE GND
RUN INTVCC
40.2k
100pF 30.1k 4.7μF S1P S1N
PLLIN Q1
16V L1
18 16 10 35 3 36 0.33μH
20.0k 47pF 25
PLLIN
ILIM
FREQ
MODE
RUN
IFAST
VIN
RSENSE1 330μF
34 INTVCC
24
0.001Ω
100μF + 2.5V
CLKOUT CLKOUT 6.3V
13 D1 CMDSH-3 Q3 Q4 SANYO
1nF VFB X5R
13.5k 33 w2
14 BOOST1
ITH
40.2k 15 0.1μF
ISET 32 TG1
TG1 VIN
– 1 31 SW1
VOSENSE DIFFN SW1
+ 2 30 BG1
VOSENSE DIFFP BG1 10μF Q1,Q5,Q9: RJK0305DPB
0Ω 38 D2 CMDSH-3 16V Q3,Q4,Q7,Q8,Q11,Q12: RJK0330DPB
66.5Ω DIFFOUT DIFFOUT LTC3829
4 26 X5R
AVP BOOST2
23
EXTVCC EXTVCC 0.1μF
17 27 TG2
PGOOD PGOOD TG2
37 28 SW2 VOSENSE+
ITEMP ITEMP SW2
39 29 BG2 S2P S2N
Q5 L2
GND BG2 VOUT 10Ω
D3 CMDSH-3 0.33μH VOUT
9 19 1.5V
BOOST3 330μF
CSS
TK/SS RSENSE2 100μF + 60A
0.1μF 0.001Ω 6.3V 2.5V
0.1μF 20 Q7 Q8 SANYO
TG3 X5R
w2
SENSE1–
SENSE2–
SENSE3–
SENSE1+
SENSE2+
SENSE3+
21
SW3
22
BG3
5 6 7 8 11 12 VIN
90
1 PHASE 3 PHASE IO
20A/DIV 60A
DN489 F04
85 100μs/DIV
EFFICIENCY (%)
VIN = 12V
VO = 1.5V Figure 4. Transient Performance without AVP
fSW = 400kHz
L = 330nH and Nonlinear Control
80 RSENSE = 1mΩ
QT = RJK0305DPB
QB = 2w RJK0330DPB
VO 166mV
STAGE SHEDDING
100mV/DIV
CCM
75
1 10 100
LOAD CURRENT (A) DN489 F02 IO
20A/DIV 60A
Figure 2. Efficiency Comparison of Stage Shedding vs CCM
22 100μs/DIV DN489 F05
VIN = 12V
VOLTAGE ACROSS SENSE RESISTOR (mV)
20
VO = 1.5V
18 Figure 5. Transient Performance with Nonlinear Control
16
14
12
10 VO
100mV/DIV 125mV
8
6
4 IO
20A/DIV 60A
2 FIRST PHASE
SECOND PHASE
0
THIRD PHASE
–2 100μs/DIV DN489 F06
0 10 20 30 40 50 60
LOAD CURRENT (A) DN489 F03
Figure 6. Transient Performance with AVP
Figure 3. Current Sharing Performance Between Phases
Data Sheet Download For applications help,
www.linear.com call (408) 432-1900, Ext. 3967
fore, their gate-to-source voltages, which in this configuration Innovation: The Reconfiguration of Existing Product Technolo-
also equal their drain-to-source voltages, are exactly equal to gies and the Failure of Established Firms,” Administrative Sci-
one-fourth of the power-supply voltage: VSG2=VSG3=VDD/4. ence Quarterly, Vol 35, pg 9, March 1990, http://bit.ly/ewuTJT.
Also note that the feedback transistor Q1 also matches
transistors Q2 and Q3, except that its aspect ratio is only one- AUTHOR’S BIOGRAPHY
fourth of their values. It also resides in its own N well, and its Eugene Bukowski is a manager and a management
source terminal connects directly to its bulk terminal. Due to consultant at Accenture for the company’s Process
the squared relationship between the gate-to-source voltage and Innovation Performance service line. Prior to
and the drain current of the MOSFET, this orientation means joining Accenture, he commercially designed analog
that the gate-to-source, or overdrive, voltage of Q1 beyond MOS ICs for both IBM and General Motors. He
VT0 must be exactly twice that of either transistor Q2 or tran- also has published 10 inventions, six of which are pat-
sistor Q3. Thus, the circuit has an operating point that must ented. Bukowski has electrical engineering degrees from Purdue
satisfy the following equation: [(VDD/2)−VSG1]=VT0. Because University (West Lafayette, IN) and Duke University (Durham,
NC), along with a master’s degree in management from Kettering
University (Flint, MI). In addition, he is a Lean Six Sigma Master
MANY ELECTRONIC DESIGNS Black Belt, including Design for Lean Six Sigma, a Shainin Red X
Master problem solver, and a senior member of the American Soci-
NOW USE ASICs WHEN ECONOM- ety for Quality.
IC INCENTIVE TO DO SO EXISTS.
any people assume that e-book readers re- they are comparing the e-reader to a paper book, not to other
M
quire only minimum processing power to electronic devices. A printed book is an instant-on device; as
render a basic text page, but this assump- soon as you turn the page, the next one is before your eyes.
tion is wrong. Displaying text may be easy, Therefore, e-readers require a high-performance processor,
but displaying it quickly—at the speed of even for text only.
paper—requires high performance. To en- Table 1 compares the classes of processors that will find
sure the wide acceptance of e-readers, they must provide a use in e-readers this year. Because of the need for low-power
user experience closer to that of reading a good-old paper consumption and compactness, all use the ARM architec-
book and to the experience that traditional consumer-elec- ture. The table shows that the recent ARM Cortex-A8 core
tronics devices offer. Today’s e-readers face these challenges: enables four times the speed of the low-end ARM9. Proces-
to be as quick as paper and as quick as LCD-based devices. sors confirm this increase. Table 2 shows the time these pro-
High-performance microprocessors, such as those operating cessor take to open a PDF (portable-document-format) file.
at 800 MHz on the ARM Cortex-A8 architecture, can en- The benchmarks use a standard LCD screen.
able better usability and new use scenarios for e-readers. One way to secure a maximum bandwidth on the CPU is
to offload it from all display-control tasks. E-reader screens
AS QUICK AS PAPER require a lot of preprocessing. The current generation uses
Go into your favorite electronics store and select two an external controller that sits between the processor and
e-readers. Press the next-page button and note how long it the display, but this controller is expensive.
takes for a page to turn. You may see a noticeable difference A next-generation chip, such as Freescale Semiconduc-
between the two readers. Page-turn time is the biggest chal- tor’s (www.freescale.com) i.MX508 e-reader processor, inte-
lenge for e-readers. Although you may tolerate the fact that grates the controller directly in the silicon, in hardware rath-
your PC takes a full second to render a page full of pictures, er than in software, to help ensure both high performance
it becomes painful when you need to wait in the middle of a and low power consumption. CPU performance has a direct
sentence, especially when this happens 300 times in a row in impact on the time it takes to render a new page, whether as
a single book. text or as an image.
It may seem paradoxical, but users expect a higher perform-
ance from a light e-reader than from a big notebook because AS QUICK AS AN LCD
Most e-readers use EPDs (electrophoretic, or electronic-
TABLE 1 ARM-BASED PROCESSOR paper, displays), and the dominant player is E Ink (www.
COMPARISON eink.com). EPDs have many benefits over LCDs for reading
e-books. For example, EPDs are bistable, which means that
Dhrystone
Type of Typical DMIPS/ performance they can hold an image without the need for updating and
processor speed (MHz) MHz (DMIPS) therefore consume no power except when turning a page.
They are also reflective, meaning that they require no back-
ARM9 400 1.06 424
(ARM926EJ-S) light, making it more comfortable for a reader’s eyes and en-
abling readability in sunlight. In short, they are like paper
ARM11 532 1.18 628
(ARM1136J-S)
(Figure 1).
This technology has downsides, however. First, EPD tech-
ARM Cortex-A8 800 2.07 1656 nology has a slow display-frame rate. Changing the color
E D
(a) (b)
ONE FULL UPDATE
LPF
LTC6409 LTC2262-14
0.9V Output
Common-Mode Set
designideas
AND FRAN GRANVILLE
VDC à R1 R5
0 TO 40V Ź 1k 150
2W
R2 Q1
10k BD139
R3
1k
J4 J3 J2 J1
IOâ0 TO 2A
VO30V
50 MUR1520 DIODES
Figure 1 You can change the photovoltaic characteristics of this simulator by modifying the current-source output value or con-
necting 10 to 50 diodes in series. You can short-circuit these diodes in groups of 10.
1.6 20
1.4 18
1.2 16
14
1 12
CURRENT (A) 0.8 POWER (W) 10
0.6 8
0.4 6
4
0.2 2
0 0
0 5 10 15 20 25 0 5 10 15 20 25 30
VOLTAGE (V) VOLTAGE (V)
30 DIODES: ISC1 10 DIODES: ISC1 30 DIODES: ISC2 30 DIODES: ISC1 10 DIODES: ISC1 30 DIODES: ISC2
20 DIODES: ISC1 40 DIODES: ISC2 20 DIODES: ISC2 20 DIODES: ISC1 40 DIODES: ISC2 20 DIODES: ISC2
Figure 2 Modifying the current-source output value also Figure 3 For each combination of diodes, the maximum
modifies the short-circuit current. power-point voltage changes.
System designers use Maxim’s field-proven technologies to reduce risk while cutting design and deployment cycles for
smart grid applications. Building on more than two decades of expertise, Maxim’s turnkey solutions are easily modified
to meet new standards and usage requirements, helping you protect your infrastructure investments.
Our IC solutions give the grid a higher IQ Smart Grid Solutions Guide
• Exceed measurement accuracy requirements Smart Grid
Solutions Guide
so your design doesn’t just meet today’s
spec but tomorrow’s as well
• Extend communications across transformers to
reach more endpoints and reduce deployment costs
• Improve power quality and delivery while
protecting assets and lowering O&M costs Edition 1, January 2011
DIRECT ™
www.maxim-ic.com/shop www.em.avnet.com/maxim
For a complete list of Maxim’s sales offices and franchised distributors, visit www.maxim-ic.com/sales
designideas
the photovoltaic-module simulator for es, modifying the current-source output REFERENCE
two current-source values and a differ- value, and, when you connect or discon- 1 Ausiàs Garrigós and José M
ent number of diodes. Extracting these nect diodes, it varies the open-circuit Blanes, “Power MOSFET is core
curves required the use of an electronic voltage of the module. You can use digi- of regulated-dc electronic load,” EDN,
load (Reference 1). The short-circuit tal circuits to control the simulator to March 17, 2005, pg 92, http://bit.ly/
current of the simulated module chang- create photovoltaic patterns.EDN fz8geK.
VISO VISO
Và VISO
R4 R2I
CONTROLLER R1I Ľ IC4
R1 X Ľ IC1 20k T1 10k
49.9k LTC6255 MOD OUT à IC3à 4.99k LTC6255 OUTPUT
PWM à 780 μH 780 μH LT1719 à
GND IC2 Và Và Ľ Ľ
C1 C1I
LTC6992-2 CS R3I
10 μF 0.22 μF
SET DIV 0.1 μF COILCRAFT 10k
RSET WB1010-SM
499k
Figure 1 This circuit creates a 100-kHz PWM signal from a 1-kHz PWM signal to send it across an isolation transformer. The
output is an integrated dc control voltage.
Figure 2 The circuit gives a gradual exponential response Figure 3 An anticipator circuit speeds the output response
to a 20 to 80% step change in duty cycle. A 500-msec time (green).
constant filters the input PWM (yellow). The filtered, 100-kHz
isolated signal closely matches the input (green).
fortunately, this pin. If you connect the fast-responding
approach would input signal to the positive input of IC6,
Idea describes (Reference 1). Adding make the response time of the recon- it would cause a large overshoot in the
the anticipator circuit at the X node of structed output slow. You cannot use the duty cycle of the output signal and a
Figure 1 results in a faster response to a anticipator circuit when you re-create long recovery time. You compensate for
final value (Figure 3). the slow PWM signal because the slow the polarity reversal by biasing the DIV
You add another voltage-controlled signal is now the dependent variable in pin of IC2.
PWM IC to re-create the 1-kHz PWM the circuit, and fast jumps in the feed- Setting the DIV pin above VSUPPLY/
signal on the isolated side (Figure 4). back voltage would result in the loop’s 2 programs the output-control polar-
You use amplifier-integrator circuit IC6 continuously hunting and never settling ity to change from 95 to 5% duty cycle
to servo-control the duty cycle of IC7. to a final value. Use instead a 10-msec with an increasing voltage applied to the
Resistor RSETI programs IC7 for 1-kHz, time-constant filter on the 1-kHz output MOD pin. An increase in the 100-kHz
5- to 95%-duty-cycle operation. The to obtain a reasonable response time and signal’s duty cycle now ramps down the
circuit forces the 1-kHz output duty then minimize duty-cycle ripple with an MOD pin and increases the 1-kHz out-
cycle to equal the 100-kHz input-signal additional lowpass network comprising put signal’s duty cycle to match it.EDN
duty cycle. Again, the supply voltage for R4I and C4I.
comparator IC5 and the PWM device Note that the feedback signal for inte- REFERENCE
must be the same. If you want minimum grator IC1 in Figure 1 is on the negative 1 “Anticipator circuit speeds signal
duty-cycle ripple, set filter R2I and C2I pin and that the feedback signal to re- settling to a final value,” EDN, March
to have a 500-msec time constant. Un- creation integrator IC6 is on the positive 17, 2011, pg 58, http://bit.ly/h7qZPo.
C3I
0.47 μF
VISO VISO
R5I VISO
10k R1I R3I
à IC5 4.99k 30.1k R4I
WB1010-SM Ľ IC6 49.9k
780 μH LT1719
SECONDARY Ľ LT6255 MOD OUT PWMOUT
R6I C1I à C4I VISO
10k 0.22 μF 0.1 μF Và
GND IC
7 CS RDIV1I
LTC6992-2 280k
0.1 μF
SET DIV
RSETI RDIV2I
787k 1M
R2I
30.1k
C2I
0.22 μF
Figure 4 This circuit re-creates the 1-kHz PWM control signal on the isolated side. It cannot use the anticipator circuit, but R4I and
C4I provide extra lowpass filtering, helping to reduce output jitter.
Make your electronics supply chain parts to one location and send-
ing them to another. “Green
green—or else behavior can be very benefi-
cial from a cost standpoint,”
egulations and the need for the removal of six materi- compliance. “The green supply says Colin Campbell (photo,
Videos, Including
Design Tips, Application
Demos, and More
Updates on Future
In-Person Workshops
and Seminars, like
this one!
Linear Technology,
American Aerospace Controls (AAC) 35 Mill-Max Manufacturing Corp 9
www.linear.com/6409
precision data-acquisition applications, Interconnect Systems Inc 23 Stanford Research Systems Inc 6
such as automated test equipment,
medical instrumentation, and profes- International Rectifier Corp 7 UBM Canon Trade Events 17
sional audio preamplifiers. The devices
EDN provides this index as an additional service. The publisher assumes no liability for errors or omissions.
feature 2.2-nV/√Hz noise density; 0.1-
I
−12 dB, so the crosstalk was better than
high-end analog stereophonic-audio systems. One of the prod-
24 dB, considering that each band was
ucts was a system comprising a 200W amplifier, a tape deck, capable of a total gain of 24 dB.
a turntable, and a 10-band graphics equalizer. The equalizer To sequentially select the analog
split the audio-frequency range of 20 Hz to 20 kHz into 10 switches, I used a CD4017 decade coun-
bands, allowing users to individually adjust the level of each ter with 10 outputs. An astable multi-
band, using vertically sliding controls, from −12 to +12 dB. In its vibrator comprising a 555 timer gener-
ated the clock signal for the counter.
stereo version, it had 10 bands for the left channel and 10 bands I used the CD4017’s Enable input as
for the right channel. The position of the sliders represented the a stop/run control for the clock, and I
frequency response. Most graphics equalizers then used one coil added a potentiometer to the 555 to
or one inductor per band, but this one used operational amplifiers vary the speed of the clock signal as a
instead of inductors. testing-speed control.
I bundled all of these electronics inside
Testing the equalizers involved feed- output needle’s voltmeter to show 12V; a box with the controls and connectors
ing the output of an audio-frequency moving it down resulted in a reading of outside. I timed myself to completely
generator to the equalizer and reading −12V. Because the stereo version of the test the equalizer. At the fastest, it took
its output on a center-zero voltmeter equalizer had 20 bands, I had to repeat 1.5 seconds per band, working out to 30
calibrated in decibels. For covering this procedure 20 times. seconds for the 20 bands, excluding about
one band of the equalizer, the audio- I first attacked the time-consuming 10 seconds to change from the left to the
frequency generator had to be set to the generation of the correct audio fre- right channel. I added 10 LEDs to provide
correct frequency range and then its dial quency for the band under test without a direct visual indication of which slider
had to be rotated to the center frequency selecting range switches and turning to move next. Within 24 hours, there was
of the band under test. I adjusted the dials. If I could rig up 10 audio-frequency no backlog, and production jumped from
DANIEL VASCONCELLOS
generator’s output to show 0 dB at the generators and then tune each one to its 10 per day to 200.EDN
output on the voltmeter if the slider of center frequency, I could then reduce the
the band was in midposition. Moving task of testing to selecting the required Kunal Ghosh is a project manager in
the slider up caused the reading on the audio oscillator for the band under test Hyderabad, India.
As a world leader in Ethernet for the Digital Home, Enables a Wide Range of Applications
Micrel is a preferred choice for attaching consumer
Digital TVs
electronics to home networks today. Featuring
energy efficient designs, with advanced physical Media Players
layer and switch technology, Micrel offers the Mobile Internet Devices (MID)
industry-leading low power line of Ethernet devices: Set Top Boxes
• PHY Transceivers Residential Gateways
• Single-port Controllers
Micrel Advantage
• Switch on a chip ICs
Green and highly integrated designs, reducing power
Micrel’s family of Ethernet devices has been adopted consumption and board real-estate
and proven by major consumer electronic OEMs around Advanced power management
the globe, with hundreds of millions of ports deployed.
Low EMI emissions and high-EMI immunity
For more information, contact your local Micrel
sales representative or visit Micrel at:www.micrel.
com/ad/DigitalHomeToday.