THE MAGAZINE OF TECHNOLOGY INSIDERS 05.09

MEMORY MAKER ROBERT DENNARD, wINNER OF THIS YEAR’S IEEE MEDAL OF HONOR, FIGURED OUT HOw TO DO DRAM RIGHT

THE 25 MOST REMARKABLE CHIPS EVER

TOUCH SCREENS GET TOUCHY-FEELY

THE NETFLIX CHALLENGE: HELP US RECOMMEND MOVIES, wIN A MILLION DOLLARS volume 46 number 5 north american 05.09

update 11 bait to track data thieves A honeypot project in Germany finds viruses and tracks down the data they steal. By Michael Dumiak

12 cellphone security threat

14 bright spots in the gloom

14 robotic seals aid research

15 touch screens with feeling

16 the rfid as

24 18 ultraviolet radios beam to life

20 the big picture Fabs get the white-glove treatment. opinion 8 spectral lines Ieee’s 125th-anniversary celebrations reveal the depth and breadth of its members’ contributions to technology.

10 forum readers discuss the state of the U.s. nuclear arsenal and the demise of the business-method patent.

27 reflections Bob Lucky’s head is in the clouds— 44 28 the computing clouds, that is. better pics: cover story Medical departments ultrasound 4 back story images will soon 48 thanks for A man vanishes. can Google find him? be much sharper [above]; a famed the memories 6 contributors 8-bit processor Robert Dennard’s 1968 invention of the one- dynamic random- 22 hands on powers a famed A high-definition cL D projector can cartoon robot access memory set the path of computing for the next half century. set you back Us $2000—or you could [top right]; For this, he will receive the 2009 IEEE Medal of Honor. By Sally Adee build your own. By James Turner Netflix’s prize adds the fuel of 24 geek life interest to the 28 the million dollar How an Apple II–era fire of genius. programming prize ended up in the head of a 30th-century robot. COVER: To improve its recommender system, Netflix put a price on its head. By Erico Guizzo & David X. Cohen daVid yEllEn By Robert M. Bell, Jim Bennett, Yehuda Koren & Chris Volinsky This PagE: ClOCkwisE fROm lEfT, Paul VOzdiC/ 25 tools & toys gETTy imagEs; solve age-old engineering problems “” Tm and 34 25 microchips © 2009 TwEnTiETh with your kids. By Sherry Sontag CEnTuRy fOx film CORPORaTiOn. all that shook the world RighTs REsERVEd; 26 careers Randi silbERman These chips unleashed groundbreaking technologies and gadgets—and In 2001, the iPod was a big hit, but are one reason that engineers don’t get out enough. By Brian R. Santo sean Adams had a different idea. 44 next-gen ultrasound By Sherry Sontag 64 the data Technology from the semiconductor industry promises to supplant the Is the high-tech unemployment glass transducers now used for medical imaging. half full or half empty? Let’s look at By Butrus T. Khuri-Yakub, Ömer Oralkan & Mario Kupnik the numbers. By Steven Cherry www.spectrum.ieee.org MAy 2009 • Ieee sPectrUM • NA 1 volume 46 number 5 north american 05.09

www.Ieee.orG/ tHeINstItUte available 6 may on the institute online tech experts speak at ieee media event Get the scoop on the latest technologies in such areas as cancer recognition, brain-machine interfaces, and cognitive computing, as discussed by seven tech leaders at the Ieee Media roundtable. the theme of the March event, held in celebration of Ieee’s 125th anniversary, was “embracing Human-technology Interactions.” networking humanitarian technology the new Ieee Humanitarian technology Network enables lEfT: maya members to connect and collaborate EnTERTainmEnT; RighT: JEffREy www.sPectrUM.Ieee.orG online with others doing similar work hamilTOn/gETTy imagEs available 1 may on spectrum online in the field. animation nation accessibility India’s burgeoning animation and visual-effects industry combines conference two great passions of India’s technorati: moviemaking and software the first Ieee conference to identify the accessibility challenges arising composition. Much of the animation and visual-effects work on feature from the pervasive use of technology films such asMadagascar: Escape 2 and Alvin and the Chipmunks, TV will be held in July in Boston. Areas shows like Nickelodeon’s “Back at the Barnyard,” and games like of focus include transportation, Gears of War is being done by Indian animation houses. But as these collaborative care, and standards. companies start to expand and create original content, their success will hinge on a combination of economic and cultural factors, many of which still remain uncertain and unpredictable, as Suhas Sreedhar discovered during a recent visit to Bangalore.

oNLINe FeAtUres: ALso oNLINe: Learn about china’s DtV • webcasts standard, which takes advantage of • Podcasts advances in wireless communication • News and information encoding to create • videos a system that works well even in bad • Blogs weather and can be viewed on the go. • Jobs • career Accelerator Got somethinG to say? • Ieee Xplore® digital library Join the conversation among our • Interviews editors, bloggers, and readers in our • opinions Contains over 25% expanded online comments sections. Renewable Resources • More!

ieee spectrum (IssN 0018-9235) is published monthly by the Institute of electrical and electronics engineers, Inc. All rights reserved. © 2009 by the Institute of electrical and electronics engineers, Inc., 3 Park Avenue, New york, Ny 10016-5997, U.s.A. the editorial content of Ieee spectrum magazine does not represent official positions of the Ieee or its organizational units. canadian Post International Publications Mail (canadian Distribution) sales Agreement No. 40013087. return undeliverable canadian addresses to: circulation Department, Ieee spectrum, Box 1051, Fort erie, oN L2A 6c7. cable address: ItrIPLee. Fax: +1 212 419 7570. INterNet: [email protected]. ANNUAL sUBscrIPtIoNs: Ieee Members: $21.40 included in dues. Libraries/institutions: $399. PostMAster: Please send address changes to Ieee spectrum, c/o coding Department, Ieee service center, 445 Hoes Lane, Box 1331, Piscataway, NJ 08855. Periodicals postage paid at New york, Ny, and additional mailing offices. canadian Gst #125634188. Printed at w224-N3322 Duplainville rd., Pewaukee, wI 53072-4195, U.s.A. Ieee spectrum circulation is audited by BPA worldwide. Ieee spectrum is a member of American Business Media, the Magazine Publishers of America, and the society of National Association Publications. www.spectrum.ieee.org MAy 2009 • Ieee sPectrUM • NA 3 back story

EditoriAl

Editor in chiEf Susan hassler, [email protected]

ExEcutivE Editor Glenn Zorpette, [email protected]

ManaGinG Editor Elizabeth a. Bretz, [email protected] Where in SEnior EditorS harry Goldstein (online), [email protected]; Jean Kumagai, [email protected]; Samuel K. Moore (news), [email protected]; tekla S. Perry, [email protected]; Philip E. ross, [email protected]; david Schneider, [email protected]; William the World Sweet, [email protected]

SEnior aSSociatE Editor Steven cherry (resources), [email protected] Wide Web Is aSSociatE EditorS Sally adee, [email protected]; Erico Guizzo, [email protected]; Joshua J. romero (online), [email protected]; Sandra upson, [email protected] Al Phillips? aSSiStant Editor Willie d. Jones, [email protected] SEnior coPy Editor Joseph n. Levine, [email protected]

coPy Editor Michele Kogon, [email protected] is name was Al Phillips, and be a she. There was an Al Phillips EditoriaL rESEarchEr alan Gardner, [email protected] journalist Brian Santo knew who’d recently been a New York ExEcutivE ProducEr, SPEctruM radio Sharon Basco aSSiStant ProducEr, SPEctruM radio francesco ferorelli, H this much about him: He’d Jets cornerback, but no way had he [email protected] been a successful semiconductor helped architect a chip in the 1970s. adMiniStrativE aSSiStantS ramona Gordon, [email protected]; executive, having founded Western Then there was the Canadian wres- nancy t. hantman, [email protected] Digital Corp. in Southern California tler, who was actually better known intErn rosaleen ortiz, [email protected] contriButinG EditorS John Blau, robert n. charette, some 40 years ago. But he seemed to as Spider Dick, and also the founder Peter fairley, alexander hellemans, david Kushner, have vanished not long after that. of a chain of dry cleaners in Hawaii. robert W. Lucky, Paul Mcfedries, Kieron B. Murphy, carl Selinger, Seema Singh, John voelcker Phillips was just one of a couple Nope, and nope. of dozen people Santo had set Santo kept on Googling, and Art & productioN out to interview for this issue’s then he got a break. A letter to the SEnior art dirEctor Mark Montgomery aSSociatE art dirEctor Michael Solita “25 Microchips That Shook the editor of the Orange County Business aSSiStant art dirEctor Brandon Palacio

World.” The idea was to get insight Journal, in California, was signed Photo Editor randi Silberman

about the chips straight from the “Al Phillips, Founder, Western dirEctor, PEriodicaLS Production SErvicES Peter tuohy designers and managers who Digital Corp., Newport Coast.” EditoriaL & WEB Production ManaGEr roy carubia created them. To reach these people, Bingo! From there, Santo turned SEnior ELEctronic Layout SPEciaLiSt Bonnie nani some long retired, Santo put out calls, to the phone book, one of the oldest WEB Production coordinator Jacqueline L. Parker sent e-mails, and, of course, searched tools of the trade in journalism— MuLtiMEdia Production SPEciaLiSt Michael Spector Google. In fact, he Googled like crazy. but not a big fat paper one. He used EditoriAl AdviSory BoArd “I wish I could just hook up Google’s phone book. Susan hassler, Chair; Marc t. apter, francine d. Berman, Jan Brown, raffaello d’andrea, Stephen L. diamond, hiromichi my brain to Google,” sighs Santo Reached at his home in Newport fujisawa, Kenneth y. Goldberg, Susan hackwood, Bin he, Erik [above, right], who started his Coast, Phillips [above, left] was only heijne, charles h. house, david h. Jacobson, christopher J. James, ronald G. Jensen, Mary y. Lanzerotti, ruby B. Lee, tak career at Electronic News, a weekly too happy to chat. But first he needed Ming Mak, david a. Mindell, c. Mohan, fritz Morgan, andrew trade newspaper, in 1984 (when, to keep an appointment at a local golf M. odlyzko, Leslie d. owens, Barry L. Shoop, Larry L. Smarr, harry L. “nick” tredennick iii, Sergio verdu, William Weihl, by the way, Google founders course. Later, when he did have time Bas¸ak yüksel

Larry Page and Sergey Brin were to talk, he told Santo he was a former EditoriAl corrESpoNdENcE 11-year-olds still getting allowances Motorola and Rockwell executive iEEE Spectrum, 3 Park ave., 17th floor, new york, ny 10016-5997 from their parents). who founded Western Digital in 1970. attn: Editorial dept. tel: +1 212 419 7555 fax: +1 212 419 7570 Bureau: Palo alto, calif.; tekla S. Perry +1 650 328 7570 After a few days, Santo had gotten He also mentioned that he’d written responsibility for the substance of articles rests upon the hold of all the people on his list. All an article published in the June 1964 authors, not the iEEE or its members. articles published do not represent official positions of the iEEE. Letters to the editor may except one, that is—Al Phillips. issue of IEEE Spectrum, “Monolithic be excerpted for publication.

An initial Google search located Integrated Circuits,” and that at age AdvErtiSiNg corrESpoNdENcE an Al Phillips in Knoxville, Tenn., 80, he’s still a busy man. He doesn’t iEEE Spectrum, 3 Park ave., 17th floor, new york, ny 10016-5997 but he turned out to be an insurance play football, but besides golf, he attn: advertising dept. +1 212 419 7760 agent. There was an Al Phillips enjoys bicycling and skiing. the publisher reserves the right to reject any advertising. who was an active member of the Google didn’t turn up those rEpriNt pErmiSSioN nto LiBrariES: articles may be photocopied for private use sA

Taftville Congregational Church in details. For some things, you still of patrons. a per-copy fee must be paid to the copyright A Connecticut, but he turned out to do need a phone. o clearance center, 29 congress St., Salem, Ma 01970. for other copying or republication, contact Business Manager, iEEE Spectrum.

citiNg ArticlES iN iEEE SpEctrum s; Right: Lis P coPyriGhtS and tradEMarKS: iEEE Spectrum is a registered i

IEEE Spectrum publishes two editions. in the international edition, the abbreviation int appears at the LL foot of each page. the north american edition is identified with the letters na. Both have the same trademark owned by the institute of Electrical and Electronics Phi editorial content, but because of differences in advertising, page numbers may differ. in citations, Engineers inc. careers, EEs’ tools & toys, Ev Watch, Progress, L you should include the issue designation. for example, the first update page is in IEEE Spectrum, reflections, Spectral Lines, and technically Speaking are trademarks of the iEEE. vol. 45, no. 5 (int), May 2009, p. 9, or in IEEE Spectrum, vol. 45, no. 5 (na), May 2009, p. 11. Left: A

 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org iEEE mEdiA Staff Director; PubliSher, IEEE SpEctrum James a. Vick, [email protected] aSSociate PubliSher, SaleS & aDVertiSing Director Marion Delaney, [email protected] RobeRt M. bell, ChRis he was vice president for recruitMent SaleS DeVeloPMent Manager Michael buryk, [email protected] Volinsky, and yehuda applied research at Telcordia buSineSS Manager robert t. ross koRen worked together Technologies, in Piscataway, N.J. Marketing & ProMotion Manager blanche Mcgurr, [email protected] at AT&T Labs Research, in interactiVe Marketing Manager ruchika anand, [email protected] Florham Park, N.J., where they sheRRy sontag liSt/recruitMent Marketing Manager ilia rodriguez, [email protected] competed for “The Million Dollar is coauthor of the rePrint SaleS +1 212 221 9595, ext. 319 Programming Prize” [p. 28]. best seller Blind DePartMent aDMiniStrator faith h. Jeanty, [email protected] JiM bennett, who was vice Man’s Bluff: The aDVertiSing SaleS +1 212 419 7760 telePhone aDVertiSing/SaleS rePreSentatiVe president for recommendation Untold Story of John restchack +1 212 419 7578 systems at Netflix, helped American Submarine Espionage aDVertiSing ProDuction Manager felicia Spagnoli Senior aDVertiSing ProDuction coorDinator nicole evans organize the competition to build (Harper Perennial, 1998). Her aDVertiSing ProDuction +1 732 562 6334 a better movie recommender. interest in digital jukeboxes led her ieee Staff executiVe, PublicationS anthony Durniak Bell is a principal member of the to purchase one made by Slim iEEE BoArd of dirEctorS PreSiDent & ceo John r. Vig technical staff at AT&T Labs and a Devices in 2003; her interest in her +1 732 562 3928 fax: +1 732 465 6444 [email protected] fellow of the American Statistical two small sons led her to try the PreSiDent-elect Pedro a. ray treaSurer Peter W. Staecker Association. Volinsky has been children’s software game Time Secretary barry l. Shoop director of the labs’ statistics Engineers. She writes about both in PaSt PreSiDent lewis M. terman research department since 2003. this issue [p. 26 and p. 25]. VicE prESidENtS Koren now works for Yahoo teofilo ramos, Educational Activities; Jon g. rokne, Publication Services & Products; Joseph V. lillie, Member & Geographic Research, in Israel. JaMes tuRneR, Activities; W. charlton adams, President, Standards Association; harold l. flescher, Technical Activities; gordon W. Day, President, author of this IEEE-USA

butRus t. khuRi-yakub, month’s Hands On diViSioN dirEctorS ÖMeR oRalkan, and MaRio column [p. 22], says giovanni De Micheli (i); robert e. hebner Jr. (ii); curtis a. Siller Jr. (iii); roger W. Sudbury (iV); kupnik can all put “Dr.” in front there’s more to a Deborah M. cooper (V); Mark i. Montrose (Vi); John D. McDonald (Vii); Stephen l. Diamond (Viii); of their names, although you can’t do-it-yourself project than just frederick c. Mintzer (ix); richard a. Volz (x)

turn to them for a prescription. building it. His new LCD digital rEgioN dirEctorS Still, the work they describe in video projector is sitting idle in howard e. Michel (1); William P. Walsh Jr. (2); William b. ratcliff (3); Don c. bramlett (4); David J. Pierce (5); leonard J. bond (6); “Next-Gen Ultrasound” [p. 44] his laundry room until he can ferial el-hawary (7); Jozef W. Modelski (8); enrique e. alvarez brings them nearly as often to clear enough wall space to display (9); yong Jin Park (10) Stanford University’s medical its capacious images. Turner is a dirEctorS EmErituS eric herz, theodore W. hissey school as to its engineering school. contributing editor for O’Reilly Khuri-Yakub, a professor of Media and a correspondent for iEEE StAff executiVe Director & coo James Prendergast , manages the Christian Science Monitor. +1 732 502 5400, [email protected] huMan reSourceS betsy Davis, SPhr technical operations for Stanford’s acko +1 732 465 6434, [email protected] l E.L. Ginzton Laboratory, where daVid yellen, PublicationS anthony Durniak +1 732 562 3998, [email protected]

Oralkan is a senior scientist and who shot the photos tephen eDucational actiVitieS Douglas gorham s Kupnik is a research associate. of our Medal of +1 732 562 5483, [email protected] StanDarDS actiVitieS Judith gorman Honor winner +1 732 562 3820, [email protected] RobeRt W. [p. 48], became a MeMber & geograPhic actiVitieS cecelia Jankowski +1 732 562 5504, [email protected] luCky ponders the photographer after working as a corPorate Strategy & coMMunicationS Matthew loeb, cae desirability of musician, fashion designer, and +1 732 562 5320, [email protected] buSineSS aDMiniStration richard D. Schwartz dispersing all your fishing crewman. When not +1 732 562 5311, [email protected] technical actiVitieS Mary Ward-callan data into the playing with his daughter, he still +1 732 562 3850, [email protected] computing cloud in this month’s fishes in Sheepshead Bay, near his Managing Director, ieee-uSa chris brantley

+1 202 530 8349, [email protected] raham; james turner; yellen; david Reflections column [p. 27]. Lucky, Brooklyn, N.Y., home. “Shooting at g he

iEEE puBlicAtioN SErVicES & productS BoArd c an IEEE Fellow, holds 11 patents IBM,” he says, “was an adventure,” Jon g. rokne, Chair; tayfun akgul, tariq S. Durrani, and worked for many years at because the Eero Saarinen Mohamed e. el-hawary, gerald l. engel, gerard h. gaynor, Marion o. hagler, Jens hannemann, lajos hanzo, Bell Labs. Before retiring in 2002, building is hidden in the woods. hirohisa kawamoto, Michael r. lightner, lloyd a. Morley, William W. Moses, adrian V. Pais, Saifur rahman, Sorel reisman, edward a. rezek, barry l. Shoop, W. ross Stone, robert J. trew, leung tsang, karl r. Varian, Stephen yurkovich

iEEE opErAtioNS cENtEr 445 hoes lane, box 1331, Piscataway, nJ 08854-1331 u.S.a.

tel: +1 732 981 0060 fax: +1 732 981 1721 clockwise fromright: top

 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org spectral lines

IEEE INNOVATORS: Engineering From left: Roy Want, Krishna Palem, Miguel the Future Nicolelis, Dharmendra Modha, K.J. Ray at IEEE Liu, Rangachar Kasturi, Katie Hall, IEEE Past nniversaries are President Lew often engulfed by Terman. Photo: IEEE A nostalgic memories of past achievements and glory days gone by. Not this one. While commemorating its members’ historic engineering accomplishments over the past 125 years, IEEE’s 125th- anniversary celebrations— going on throughout 2009 around the globe—focus environmental sensing. waves. Krishna Palem, an “Thanks for the Memories,” on how today’s engineers K.J. Ray Liu, an IEEE IEEE Fellow and a professor in this issue), described his are inventing and building Fellow and a professor at at the George Brown School own work on one-transistor technologies that will shape the University of Maryland, of Engineering at Rice dynamic RAM like this: “We the world’s future over the College Park, discussed the University, reported on his just didn’t imagine how far it next 125 years. development of predictive work on low-energy chip would go, how much it would Take, for example, the cancer testing—tests that technology. He’s now using totally change computing.” media roundtable IEEE would be able to identify it in a low-power, low-cost So if you have students, put on in New York City both patients whose cells are “I-Slate” that will give children point them to the round- in March to showcase the transitioning from normal in impoverished areas— table webcast so they can work of just seven of its to cancerous and the type who may have never even seen experience the magic members (you can still view of cancer that’s developing— a teacher—access to learning. of watching people talk the whole presentation at based on algorithmic analyses Finally, Roy Want, an IEEE passionately about the work http://www.ieee125.org/ of genome-proteome signaling. Fellow and senior principal they love. If you’re a student, engineering-the-future/ Dharmendra Modha, an IEEE engineer at , talked watch the webcast and listen media-roundtable.html). Senior Member and manager about a mobile solution carefully. New fields and new It’s well worth checking out. of cognitive computing at called dynamic composable professions will be spun out First, Katie Hall, an IBM’s Almaden Research computing that could make of the work in power, robotics, IEEE Senior Member and Center, related his team’s it possible for you to access computing, and biomedical chief technology officer of work on SyNAPSE, a project any devices or applications engineering described therein. WiTricity, talked about a to build a “brain”—a computer you need on the fly—through You will learn, if you haven’t magnetic resonance–based system that can see, feel, and your cellphone. already, that a universe of technology her company think like a human being. Phew! Clayton M. opportunities to invent new is developing that can In another intriguing Christensen coined the term technologies and make a wirelessly transmit power, biomedical engineering disruptive innovation, and if difference in the world remain from milliwatts to kilowatts, development, Miguel Nicolelis, some of these developments wide open to the curious to electronic devices meters an IEEE Member and don’t turn out to be disruptive and adventurous willing away. Rangachar Kasturi, an professor and codirector of the I don’t know what will. But to take chances and follow IEEE Fellow and a professor Center for Neuroengineering then again, sometimes new their dreams. —Susan Hassler at the University of South at Duke University’s Medical technologies take their own Florida, described his team’s Center, talked about a chip his sweet time to reveal their For more information about work on pattern recognition group has built that allows power and influence. This IEEE’s 125th-anniversary and its uses in medical image monkeys to control robots year’s IEEE Medal of Honor celebrations, go to http:// analysis, biometrics, and across continents with brain winner, Robert Dennard (see www.ieee125.org.

 NA • iEEE SpEctrum • mAy 2009 WWW.sPEcTRuM.IEEE.oRg forum

a period of increasing PATENT failures. The end of PROTECTION life for such a system enjoyed Steven J. is defined as the point IFrank’s “The Death where failures begin of Business-Method increasing. However, Patents” [March]. The modern reliability Bilski decision is a start thinking posits that there toward curtailing these are at least six patterns insidious bits of patent of failure, of which the overreaching, but we bathtub curve is only have far to go. I also think one. In fact, it has been something more basic shown that as systems is involved, something become more complex, to do with adjusting to two curves predominate. the change from the Both are characterized by industrial age to the the lack of a determinate . For end-of-life stage. example, our present The stockpile should economic troubles are, I Letters do not STOCKPILING off maintaining designs be replaced with more would argue, a product of represent opinions SECURITY that we know work than modern equipment, but the information age. With of the Ieee. short, concise letters are s a physicist and we are fielding untested for reasons other than the aid of the computer, preferred. they Aengineer with over ones. However, I would reliability. Our current financial engineers may be edited for 36 years’ experience argue against halting nuclear arsenal was created products and space and clarity. in nuclear weapons new developments. The designed to deliver large- processes that few Additional letters are available online laboratories, I found book on nuclear weapons yield weapons to targets understood and that led in “And More Forum” “What About the Nukes?” design is far from closed. in the Soviet Union. to an artificial, temporary at http://www. [March] well reasoned. For that reason alone, the Today the threat from increase in financial spectrum.ieee.org. At the same time, I have United States would be the former Soviet nations assets. Incidentally, Write to Forum, some issues with its well advised to maintain is minimal, and we have many of these financial IEEE Spectrum, 3 Park Ave., 17th conclusions. The ability a vibrant nuclear weap- the ability to precisely products and processes Floor, New York, NY to design a fundamentally ons research, design, and target our weapons, were patented. 10016-5997, U.s.A.; new weapons system development program. obviating the need for David S. Holland fax, +1 212 419 7570; without at least a such large warheads. IEEE Member e-mail, n.hantman William J. Camp Alexandria, Va. @ieee.org. minimal testing program IEEE Member Another reason to renew is at best suspect. The Cedar Crest, N.M. the arsenal is the need to answer to this problem train a new generation of The author responds: The as promulgated by one have been an electri- weapons engineers. The urgent issue is to not of the creators of the Ical engineer in a process maintenance of 50-year- throw the baby out with Stockpile Stewardship industry for more than old legacy systems may the bathwater: We don’t Program, Victor H. Reis, 38 years. I don’t have any not be enough to attract want industrial-age is to rely on simulation. background in nuclear the best and brightest. prejudice to preclude all I was one of the initial weapons engineering, That by itself could lead information-age patents, laboratory directors of but I have had exposure to failures of the system. and if Bilski is taken too the stockpile stewardship to reliability engineering literally, software will John M. Briggs simulation program, the concepts, and I believe IEEE Senior Member become off-limits almost Accelerated Strategic that these concepts are Vancouver, Wash. entirely. If a patent Computing Initiative, quite broad in their claim requires even a and I take great pride application. Your article ow many nuclear modicum of technology, in its accomplishments. implies that weapons Hwarheads does one it should pass muster as Nonetheless, I am skep- systems follow a country need? 5000? 500? patentable subject matter. tical that the simulation “bathtub” curve, where 50? 5? Most countries get That doesn’t mean a agenda is sufficiently there is an early period of by with none. patent will ultimately be mature to be relied on infant mortality, followed awarded, just that the Peter Flanagan as a full alternative to by a long period of IEEE Member claim is eligible for con-

testing. We are better stable performance and Melbourne, Australia sideration on the merits.

10 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org more online at www.spectrum.ieee.org

the seven or eight current varieties CatChing of “keyloggers”—malware that Crooks: Setting Bait records keystrokes in order to Honeypots can tell security pick up passwords and other researchers how to Track Data Thieves sensitive data. The two families data is stolen were Limbo/Nethell, which uses and reveal where Germans use unguarded PCs as honeypots sham Web sites to infect visitors cybercriminals hide their lured by spam or other social- stolen goods. engineering tricks, and its more photo: John Lund/Sam diephiuS/getty imageS even hum the Internet. By leaving themselves sophisticated cousin ZeuS/Zbot, through the Rhineland unguarded and pretending to be a keylogger that uncoils from Snight at the University operated by naive humans, they an attachment and hides in the of Mannheim’s Laboratory for tirelessly troll for the latest in spam, user’s browser to steal passwords Dependable Distributed Systems. worms, viral infections, and mal- and account information. All they do is collect bad news ware. Then the honeypots execute The team followed the and nasty infections from the the bad code and record what Mannheim honeypot data to open Internet. happens. Researchers hope that open “drop zones,” computers This is the lab’s honeypot by studying the results they’ll get a where cybercriminals network, says Thorsten Holz, a better understanding of how data compile and aggregate doctoral student at the lab. The is stolen and what happens to it. their ill-gotten gains before honeypots are machines that are Holz’s team recently reported transferring them elsewhere. walled off from the German uni- the results of a seven-month study In analyzing the drop versity’s network but connected to that focused on two “families” of zones, the team found that:

www.sPeCtrum.ieee.orG May 2009 • IEEE SpEctruM • Na 11 “there isn’t a world water crisis—yet.” WIllIaM coSgrovE, coordinator of the united Nations’ latest report on the state of the world’s water

• The drop zones housed average time a computer Security firms use 33 gigabytes of account data remains infected honeypots to identify new with either of two types from 177 000 compromised 2 Days of keylogging malware malware they must defend machines in 175 countries, their clients against. But including 10 775 unique according to Roesch and bank-account credentials. Using honeypots to the big communications Yuval Ben-Itzhak, chief • Facebook is the most analyze malware came networks run by BBN technical officer of the Web popular site for stolen into its own about a Technologies and GTE. security company Finjan, social-network credentials, decade ago through the “What’s amazed me the the kind of half-year- Windows Live for work of security pros like most is the explosion in the long, in-depth analysis Web-mail user names and Lance Spitzner, author of different ways they’re used,” the Mannheim group did passwords, and eBay for Honeypots: Tracking Hackers Spitzner says. “There’s takes too long to be done online trading accounts. (Addison-Wesley, 2003); client honeypots, [Voice by most security firms. • ZeuS can parse account Marty Roesch, creator of over Internet Protocol] The Mannheim lab’s information to read bal- the popular Snort intrusion- honeypots, Bluetooth next project is to further ances (the mean value was prevention and detection honeypots. Ten years ago automate the honeypots. US $1700, the average $5225). software; and Ron Gula there were really only one “We have automated Worrisome as the results (formerly of the National or two choices.” Spitzner collection, analysis, and are, the honeypots themselves Security Agency), who founded the Honeynet monitoring. The next step is are a step forward in under- started working with Project in 1999, and it now to do this on a larger scale,” standing black-market data. honeypots in order to defend has worldwide presence. Holz says. —Michael Dumiak

The Mobile Infections Threat Why haven’t smartphones been overrun by a deluge Bluetooth virus of viruses the way 500 minutes 650 minutes 800 minutes PCs have? According to researchers

at Northeastern ity University, in Boston, S niver these phones have u tern been safe because no multimedia S message virus single mobile-phone 25 minutes 40 minutes 45 minutes orthea n

has h/ virus origin: infection rate: 0% 100% C

achieved a big enough ear S e market share for a major virus outbreak to occur. But that could change. The Northeastern r group used anonymous call data from 6.2 million mobile-phone subscribers to simulate etwork n

the epidemiology of a virus outbreak. Like flu and SARS, Bluetooth viruses spread ex slowly, through close physical proximity [top]. On the other hand, viruses embedded in L multimedia messages such as picture messages spread much more rapidly [bottom] but are confined to groups of people who know each other and share the same kind of phone. Center for Comp

12 Na • IEEE SpEctruM • May 2009 www.sPeCtrum.ieee.orG some Bright spots in the gloom he technology industry is suffering mightily from the global economic crisis. Worldwide semiconductor revenue is expected Tto drop 24.1 percent in 2009, to US $194.5 billion, according to Gartner, a technology research firm in Stamford, Conn. , Robotic Baby Seal revenue from enterprise software—that is, corporate-scale systems— will be flat at about $222 billion. Still, there are some sectors that will Could Diminish Dementia gain, some bright spots in the gloom. —Samuel K. Moore It’s soft, it’s cute, it’s cuddly, and it’s powered by two 32-bit reduced-instruction-set-computer us $7.15 $7.36 leds: the LCD tV market may be all gloom and

8000 1500 billion billion doom, but LeDs are increasingly3.0 becoming the way to . 7000 light the screen up. they replace cold-cathode fluorescent 2.5 1200 Paro may look like a toy, 6000 lamps, leading to slimmer tVs or screens with an

2.0 5000 improved900 contrast ratio. the LCD tV market will gobble up but it’s quickly attracting the

4000 $163 million worth of Le1.Ds5 in 2009, according to market serious attention of rehabilitation 600 3000 research firm si uppli, in el segundo, Calif. that’s more 1.0 2000 than double last year’s figure. By 2012, LCDt V makers researchers. 300 0.5 1000 will be spending $1.4 billion on them. manufacturers are In , more 0 using0 LeDs more often to0.0 backlight LCD tVs because of 2008 2009 efficiency improvements and the development of lower- than 1000 WorldWide cost manufacturing in taiwan and south Korea, says units have been led revenue Jagdish ribello, isuppli’s LeD analyst. even more gains will surely come, now that engineers have figured out a way to sold to care fight an efficiency-sapping phenomenon callede L D droop. providers in

$2.7 nursing homes

3.0 billion $1.9 in a slump, the drive to do and hospitals, 2.5 virtualization: billion 2.0 more with less is acute. so virtualization software, which as well as to feed me! paro gets 1.5 basically lets a company get more out of its computing

1.0 its charge from a plug infrastructure by creating the equivalent of multiple consumers who 0.5 that looks like an 0.0 machines per computer, is a natural fit. Gartner says the want a robotic 2008 2009 infant’s pacifier. overall market will grow 43 percent to $2.7 billion in 2009. photoS: top, piCture ContaCt/aLamy; WorldWide companion. Bottom, roByn BeCk/afp/ the largest component of that will be the $1.3 billion getty imageS virtualization- worth of virtualization management software Gartner softWare Short-term revenue expects firms to sell this year. experiments in Japan and the United States show that Paro can mems: open up 1 in 10 new cellphones shipped in have positive effects on the mental 2008 and you’d find a mems accelerometer, according $1.37 $1.22 billion to isuppli. And mobile phones will continue to drive health of some elderly people. Now 1500 billion 3.0

2.5 1200 growth this year—not just for accelerometers but

2.0 long-term studies are under way 900 for zoom and auto-focus actuators, pico projectors, 1.5

600 1.0 in Europe. The results could lead 300 gyroscopes, and rF filters, too. together with consumer 0.5 0 2008 2009 items0.0 like game controllers and digital cameras, mobiles to specialized versions of Paro to WorldWide will push the market up 12 percent in 2009, to nearly mems $1.4 billion. mems manufacturers are making it easier help specific groups of people, such revenue to include accelerometers. For example, in march, as elderly individuals suffering stmicroelectronics unveiled an accelerometer for handhelds that can be hooked directly to a mobile’s from dementia or children with battery without an intervening voltage regulator and that autism. For more, see http:// works consistently even as the battery runs down. www.spectrum.ieee.org/may09/ moreparo.

14 Na • IEEE SpEctruM • May 2009 www.sPeCtrum.ieee.orG a factor of 10. The higher the amplitude of oscillation, the lower the coefficient of friction. When you run a finger across the TPaD, “you get a very strong tactile sensation of something being there, like a bump, a dip, or an edge,” says Peshkin. To news create the feeling that you’re briefs rubbing your finger against a file grating, for example, you turn eleCtriC the plate oscillation on for the tomato grooves and off for the ridges. A Purdue University The entire plate vibrates, so food scientist the amount of friction is the same has designed all over the TPaD’s surface at a device that any given time. But because the generates a room- oscillations are modulated as temperature your finger’s position changes, plasma field the device fools you into thinking inside a sealed that there are varying amounts food package. of friction at different locations. The plasma makes ozone The prototype uses optical that kills E. coli touch screens sensors to keep track of your and other finger’s position. The friction unfriendly reduction can be switched on bacteria. The With feeling ozone eventually and off so quickly (within about engineers add texture to touch-screen devices reverts to 4 milliseconds on average) that ordinary oxygen. the pitch of virtual bumps or photo: tom CampBeLL/purdue dips can be made far finer than univerSity ne of the most sought- Pattern Display, or TPaD. It can what a fingertip can discern. after new features on create the illusion of texture on The group is currently O mobile devices is the an unadorned piece of glass. working on several engineering touch screen. But that name is a The 25-millimeter-diameter challenges, including creating misnomer, according to a group prototype takes advantage of larger versions of the TPaD. of researchers at Northwestern the relatively high coefficient of They’ve already gotten it to University, in Evanston, Ill., who friction between human skin work on screens comparable point out the obvious—that these and glass. “Glass is remarkably in size to those on cellphones. displays offer minimal tactile ‘sticky,’ ” says Peshkin. “You Another technical hurdle is lim- feedback. But suppose touch think of it as being smooth, iting how much power the TPaD screens could touch you back: but the coefficient of friction draws, which is hugely impor- You could “feel,” say, the edges of between glass and your tant in mobile devices, where your the buttons on a virtual keypad fingertip is about 1.” (That’s power budget is measured in milli- or the links on a Web page. about the same as rubber on watts. “In principle, the TPaD Recently, at the IEEE- dry concrete.) But when the does not need a lot of power,” says sponsored World Haptics glass is vibrated ultrasonically, Peshkin. “But in practice, you run

m Conference 2009, in Salt Lake City, a cushion of air forms between into issues like parasitic losses i C he Michael A. Peshkin and J. Edward your finger and its surface. because of power being dissipated L tC Colgate, mechanical engineering The TPaD’s 1.6-mm-thick through things like mountings.” herevkoff/ professors and codirectors of glass layer is set into ultrasonic Asked when an online shop- Northwestern’s Laboratory for oscillation by a piezoelectric per can expect to use a TPaD to

g Intelligent Mechanical Systems, ceramic disc glued to the glass. tell how a garment feels, Peshkin etty etty described their candidate for The amplitude of the oscillation said, “That’s getting into haptic i mage such an enhanced touch screen— can be controlled, reducing the virtual reality, which is a ways S

a device they call the Tactile coefficient of friction by up to down the pike.” —Willie D. Jones

www.sPeCtrum.ieee.orG May 2009 • IEEE SpEctruM • Na 15 RFID Chips Gain Computing Skills one way to do long computations with short bursts of power

n the 2000 movie Memento, things—appliances, cars, the main character tries smart clothes, and so on. I to solve the mystery of “We’re working on his wife’s murder, despite software to make it possible suffering from amnesia that to actually compute, given causes his brain to effectively that our power is going “reboot” every 5 minutes. to be disappearing and In the world of computing, reappearing,” says Kevin Fu, “passive” radio-frequency assistant professor of com- identification (RFID) chips puter science at the University have a similar problem. of Massachusetts Amherst. Dependent for power on Mementos does two infrequent, scavenged RF things: It makes sure that the energy from a reading device, RFID keeps working toward smarter than it looks: rFID chips can do complex RFID chips may reboot more finishing a computation, and calculations, thanks to Kevin Fu’s software. photo: Ben ranSford than once per second and it also keeps the chip in a state then lie dormant indefinitely, such that if it loses power, it the real world with com- and compute; and active (2.2 waiting for the next reader to can quickly resume work puting? I think what Kevin to 25.2 mW) in which it can come along. when the power returns. and his guys have done is an even write to flash memory. But computer scientists in One way the software does advance in that direction.” “As much as possible we’d Massachusetts are working that is to have the chip Fu and his colleagues like to protect programmers on software, aptly named perform energy-intensive developed their RFID from the underlying Mementos, that could tasks, such as writing data computing software on problems” of fluctuating allow an RFID to perform to flash memory, only when Intel’s prototype Wireless power and regular reboots, computations that span many ample power (more than Identification and Sensing says Benjamin Ransford, a power losses and reboots. The 2.2 milliwatts) is available. Platform (WISP), a postage- graduate student in Fu’s lab. software may enable the chip Ravi Pappu, cofounder stamp-size RFID chip with Jason Flinn, associate to compute cryptographic of the RFID company specs suited for a 1980s home- professor of computer protocols, leading to more ThingMagic, based in brew computing enthusi- science and engineering at secure signals. And it might Cambridge, Mass., says ast: a 16-megahertz micro- the University of Michigan, allow RFID chips to be more that the work of Fu’s processor, 512 bytes of RAM, says he’s impressed by than just data collectors. team is very important. and 8 kilobytes of storage (in Mementos but thinks that They could analyze and “We have millions of the form of flash memory). it still has a long way to go. possibly take action based computers everywhere, but The amount of RF power The University of on changes to the stress on computers have been chained WISP picks up, Fu says, Massachusetts work “asks a “smart” bridge or to trends to desks and server racks and can vary drastically. And some very interesting in a person’s vital signs, for other kinds of infrastructure,” Mementos must make pro- questions and has some instance. Such computational Pappu says. “Rather than grams run on WISP perform preliminary ideas,” he RFIDs could play a role in being constrained by drag- under each scenario—sleep says. “But I don’t know that the transformation of the ging computers of various (0.2 to 2.5 microwatts); mid- they’ve solved those problems Internet from a network of shapes and sizes into the range (1.8 to 3.6 mW), which or fully validated it yet.” computers to a network of real world, could we equip allows it to read from memory —Mark Anderson

16 Na • IEEE SpEctruM • May 2009 www.sPeCtrum.ieee.orG cubic meters per year of freshwater to be generated by desalination technologies in 2020, 54 BIllIoN according to lux research, in Boston. that’s about one-third of what flows out of the Nile river.

news briefs

bug batteries MIT materials scientists say bat-signal: ultraviolet they’ve used ultraviolet radios transmitters beam their signals genetically into the sky [green]. receivers look engineered to the scattered uv light [gray]. viruses to image: iStoCkphoto Beam to life construct electrodes secret military communication scheme of the lab—the transmitter was for lithium- from the 1960s is finally practical a massive laser, and the receiver ion batteries. was a bulky vacuum-tube-based The viruses, photodetector. But thanks to which infect bacteria, were he U.S. military has been range wireless links, such as in materials-science advances, the designed to grow chasing ultraviolet (UV) unattended ground sensors. The new transmitters are tiny, com- amorphous iron T communication for decades. U.S. military is interested in such mercially available UV LEDs. The phosphate along Now researchers say radios that short-range communications receivers have also shrunk to tiny, their surfaces and then bond communicate using UV light are because they can’t be intercepted solid-state avalanche photodiodes— with carbon finally within reach. Working or jammed outside their intended devices in which a single photon nanotubes. with the Army Research Lab range. What’s more, within its produces an avalanche of electrons. The result was (ARL) in Adelphi, Md., these limited range the UV-C band has With currently available devices, a nanoscale researchers are mapping out the an inherently high signal-to-noise and under typical operating con- structure with charge and steps needed to commercialize UV ratio, enabling the use of very- ditions, a low-power UV-C system discharge rates radios. They’ve reached the last low-power transmitters, accord- with the right overlap could trans- comparable piece of the puzzle: untangling ing to ARL scientist Brian Sadler. mit roughly 100 kilobits per second to those of the poorly understood, extraor- In contrast with other optical at 10 meters, dropping to less than electrodes in state-of-the-art dinarily complex way ultraviolet schemes, which rely on the 10 kb/s at 100 meters, still more Li-ion batteries. light scatters. If they can do that, transmitter sending a signal more than enough for good digital audio. photo: mit they will have unlocked the secret or less directly to the receiver, a A major remaining challenge is to a new form of non-line-of-sight UV system can take advantage of modeling the behavior of the signal communication. the signal scattering in the atmo- as it scatters randomly in the sky. Proposed UV radios commu- sphere. The transmitter beams a “It’s not like there’s a mirror nicate in the so-called solar blind modulated signal into the sky in up there,” Sadler says. At the portion of the UV-C band—light the shape of a cone (think of the University of California, Riverside, having wavelengths from 200 to Bat-Signal). The receiver is trained he and electrical engineering 280 nanometers—which, unlike on the sky as well, at an over- professor Zhengyuan Xu the sun’s UV-A and UV-B rays, lapping angle. That positioning experimented with different is almost completely blotted out makes it ideal for sensors in dense transmission sources and by the atmosphere. Near Earth’s urban environments where line-of- receivers to characterize the angle surface, even a strong UV-C signal sight communication doesn’t work. at which the transmitter beam would die off within a few kilo- “This goes around corners, through and the receiver’s field of view meters, as individual photons forests, anyplace you can get light,” should cross. “It’s not just about are picked off one by one by oxy- says Russell Dupuis, an electro- getting the overlap volume as large gen, ozone, and water molecules. optics professor at Georgia Tech. as possible,” Xu adds. “When it’s But that attenuation also makes Early UV-C radio prototypes narrower, the signal is sometimes UV-C radiation ideal for short- were far too clunky to make it out more enhanced.” —Sally Adee

18 Na • IEEE SpEctruM • May 2009 www.sPeCtrum.ieee.orG 20 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org the big picture

A FAb LiFe No, it’s not a butler presenting the lord of the manor with stud and cuff-link options. It’s a clean-room technician holding a polymer film disc that’s been embossed with dozens of microchips using Jenoptik’s microscale thermoforming process. Jenoptik says its hot-stamping technique can cheaply and reliably burn in nanometer- scale features to create chip- scale biomedical laboratories. The fully automated process can yield dozens of different chip configurations, including microscopic scaffolding for cell cultivation and tiny reaction chambers prewired with microelectrodes. Image: Jan-Peter KasPer/dPa/Landov hands on

1 2

6 7 8

The ProjecTor lighthouses.) It doesn’t do a good job plans for a generic 15-inch display-based of projecting light, though, so you also enclosure. The entire package, including ProjecT need a different kind of compound lens, shipping, ran just under $450. Build your own digital known as a triplet lens. I picked up a refurbished Samsung movie projector Here, the first Fresnel lens beams light 15-inch LCD monitor for $124 on from a source through the liquid-crystal eBay. The Samsung had been rated good high-definitionLCD display, which I’d stripped from an LCD by DIYers as very easy to strip, and projector can still set you back panel. On the other side a second Fresnel sure enough, it took me less than A US $2000, but if your wiring and focuses the image to a point on the third 30 minutes to do it. I overbought on woodworking skills are up to speed, you lens, which projects the image out onto a wood, probably using about $50 worth can put one together for around $600. plane about 3 meters away. of high-quality 2-centimeter plywood. It’s not like you have to build it from That’s the heart of the project; The trickiest part was getting the scratch. At least two companies, DIY the rest is just the casing. The bulb light-path measurements right so that for Life and DIY Projector Kits, sell kits gets quite hot, so a Lexan or tempered everything lined up. Then there was the and supplies that make this something glass heat shield is placed right before adjustable focus ring. The instructions any engineer could do in a weekend. the innermost lens. A fan at the back in the kit suggest starting with a 5-cm The key items are an LCD panel pulls cooling air down over the LCD shower drain. Perhaps drains differ and a pair of Fresnel lenses. A Fresnel panel from a slit in the cover. from region to region, but after a lot of lens is thin and light, made out of a I bought nearly everything from DIY carving with a Dremel hand tool and piece of plastic into which concentric Projector Kits: the lenses, heat shield, some plastic fumes inhaled, I finally did grooves have been cut, giving it a large high-intensity light and ballast, a thermo- have a pretty nice adjustable focus. aperture and short focal length. (The stat to run the fan, some hardware to I mounted the lens with a 1.25-cm first Fresnel lenses were invented for mount the bulb, and a set of measured vertical offset, to provide ventilation,

22 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org LIGHT SHOW: the heart of the digital movie projector is an LcD panel [1], which first needs to be stripped from its casing [2]. A triplet lens is mounted at one end of what will be the projector’s wooden case [3]. the LcD panel is mounted in the middle. Note the lens and bulb positions marked 3 4 5 [4, 5]. Here is the projector’s interior, seen from the side, front, and rear [6, 7, 8]. When com- pleted, the rear will house the bulb, bal- last, and a thermo- stat for the fan [9], while the center will have the two Fresnel lenses, the LcD panel and con- trol circuitry, and a heat shield [10]. At last! Lights, projec- tor, action [11]! 9 10 11 photoS: JameS turner

and attached the bottom and right- A few words of advice: Experience leaks. And, of course, I’ll need to side panels. Next I cut rear holes with optics isn’t required but would be replace the cracked LCD panel. for the fan and plug/power-switch helpful. And besides the electronics, During testing, the projector threw module and mounted them. Then which involves little more than basic a very nice wall-filling image from my I attached the rear panel, cutting wiring, you’ll need to be comfortable MacBook’s Digital Visual Interface port. I grooves into two pieces of wood to working with wood. Nothing was don’t have the equipment to measure the space the middle lenses, the LCD particularly difficult, but the measured output, but 1500 to 1700 lumens is com- panel, and the heat shield the correct plans leave a lot for you to figure out. monly reported for DIY projectors of this distance from the front lens. Finally, You should also be aware that at type. By contrast, a high-definition pro- I wired the electronics and the power the end of the day you’ll end up with jector from Best Buy will put out around to the LCD panel. a large, heavy appliance—a piece 1200 lm. It will also cost hundreds of dol- I tested the optics before adding of furniture, in fact. The version I lars the first time you have to replace a the right panel, with more grooves to built measures 76 by 43 by 28 cm and bulb, which you’ll have to do every 2000 hold the tops of the lenses and the LCD. weighs around 10 kilograms. Also, to 4000 hours. The one I used should While I was installing the LCD panel, I unless the LCD panel can handle last around 10 000 hours, and it cost $50. twisted one of the mounts, cracking the video input, you’ll have to drive the Now if I only had a wall in my house big display. It still functioned well enough display from a computer. There’s also enough to use it with! An ideal projection for testing. no zoom—you control display size by area would be around 2.5 meters, mea- Finally, I put the top on, temporarily moving the unit. sured diagonally. —James Turner without a ventilation slit or hinges. The project took me about 25 hours I attached my MacBook to the LCD over the course of two weeks. I still DIY For Life: http://lumenlab.com panel input and fired up the lamp. It have to hinge the cover, put in the DIY Projector Kits: worked perfectly the first time! ventilation slit, and seal all the light http://diyprojectorkits.com

www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 23 geek life

a host of other systems that fomented I knew I was in for bad news when the the personal computer revolution. return letter they sent me started with a The chip’s cameo on “Futurama” misspelling of my name. rocked the nerdosphere, prompting a From a programming perspective, a burst of commentary in online forums more impressive feat was the creation of (the nerd equivalent of cocktail party an actual working computer language chatter). There was also a description (“FLEET”) for the Apple II Plus that on Wikipedia, in the site’s entry I developed with two high school about the 6502. But we here at iEEE friends, David Borden and David Spectrum had a few questions that Schiminovich. We called ourselves somehow didn’t come up amid all the “The Glitchmasters.” This language noise. For starters: Why the 6502? was intended to make it easy to write And also: Is it possible that ’s high-speed graphics programs (that maker, Mom’s Friendly Robot Co., is, video games) for the Apple. None of somehow obtained Bender’s design us knew anything about compilers, yet from a scruffy-bearded, sandal-wearing using no references on the subject and computer geek who lived in northern working entirely in 6502 assembly, we California circa the late 20th century? somehow wrote a working compiler. To get to the truth, Associate Editor It is even more impressive when you Erico Guizzo sought the brain behind consider that virtually no comments Bender’s brain. That’d be David X. Cohen, appeared in the program—just page The TruTh AbouT executive producer and head writer of after page of assembly language. bender’s brAin “Futurama,” who, as it turns out, is quite a In fact, the resulting compiler David X. cohen, of geek himself. Here’s Cohen’s response: was extremely good: It was lightning “Futurama,” reveals how fast, the language was easy to learn mos technology’s 6502 spent a good percentage of my and program, and the compiled processor ended up in the high school years programming the programs were comparable in speed Apple II Plus in 6502 assembly language, to anything we would have written robot’s head I so I have fond memories of long nights directly in assembly language. I believe alone with this chip. My greatest 6502 it would have been an extremely On 14 November 1999, an episode of achievement was a video game I called useful product. However, our timing “Futurama,” the animated sci-fi comedy Zoid that was played heavily by me and was astoundingly poor. The compiler series conceived by “” my father and no one else. Incidentally, was completed in 1984, just as the erved. S e

creator , jolted computer Zoid incorporated digitized speech (me Apple II was fading forever into r

geeks with a display of technological saying the word “Zoid,” slowed down to oblivion and we were heading off to S ight acumen absolutely unprecedented make it mightier), which was pretty rare college. Thus our fabulous compiler r ll ll in prime-time entertainment. In the at the time. The digital audio for that never really got used for anything. a episode, “Fry and the Slurm Factory,” a single syllable used much more memory In retrospect I would say the character named than the entire program. I tried to sell limitations of the 6502 forced us, against points his F-ray at the head of the the game to Broderbund Software, but our wills, to be clever and learn the show’s famously ill-tempered robot, workings of things at a deeper level. Bender. It reveals a little rectangle, For example, we had to write our own apparently a chip, labeled “6502.” efficient subroutines to multiply and The 6502 was a beloved—at least by divide 16-bit numbers, using only 8- geeks—8-bit microprocessor created bit addition and subtraction and bit wentieth Century Fox Film Corporation. Corporation. Film Fox Century wentieth

by MOS Technology in 1975. It was the shifting. As another example, it is t chip that the scruffy-bearded, sandal- possible (in fact trivial) in computer wearing Steve Wozniak used to build graphics to compute the pixels along

the Apple II in 1977—“The Machine That a line segment from (A, B) to (C, D) 2009 © and tm Changed Everything,” as pc World without using division or computing ma” ma” once put it. It was also used in the the slope. Again it requires cleverness, a

Commodore PET, the BBC Micro, and though. So I think programming “Futur

24 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org the 6502, especially in the days While I can claim responsibility yes, in all seriousness, Ken confirms of limited , was for the appearance of the 6502 in that he does read every issue of IEEE very useful in terms of learning to “Futurama,” I was not the most highly Spectrum and occasionally looks at think creatively and efficiently. trained computer scientist or engineer Transactions on Information Theory). Moving now 15 years into the on the “Futurama” writing staff. I have No doubt Woz’s head still survives future to the year 1999, I was working a master’s degree in computer science in a jar in the year 3000, and somehow on an early episode of “Futurama.” from the University of California, it is probably wearing sandals. So Bender was being X-rayed (actually, Berkeley. However, writer it is quite possible that he provided “F-rayed”), and we needed to see has a Ph.D. from Harvard in applied Bender’s design to Mom’s Friendly what was powering his mighty math, as well as a master’s from Robot Co. in return for some extra fish robot brain. Naturally, the 6502. Stanford in electrical engineering (and food in his jar. o tools & toys

Time WAiTs Shingler has solved all at one point passionately Time Engineers US $19.95; Soft­ those puzzles with Time insisting: “We are in the ware Kids, Valparaiso, Ind.; http://www. timeengineers.com imageS: SoFtwareKidS for no Engineers, a video game in desert. Sand is cheap!” engineer which players ride a magnetic I argued for clay as a A new computer egg back in time to three more cost-effective, sturdier hundreds of high school game lets kids solve eras. In each, players solve ramp material. It would teachers around the country age-old engineering two engineering problems. have less give and require in their classrooms and by problems As would-be electrical fewer workers to pull stones thousands of parents—lots engineers, they create up. Point one to Mom. of them engineers who ow would you have radar arrays; as mechanical Coby did, however, work want to introduce their kids arranged a radar array engineers, they design a out faster than Mom how to the family business. H to protect London diesel submarine; and as civil wide a radar sweep we Sure enough, Coby now during the Blitz? How do you engineers they build those needed to save London from thinks engineering is cool, lower a drawbridge without drawbridges and pyramids. an air attack. I nailed the even though Time Engineers crashing it, using only Older kids can click to number of engines and the is not perfect. It’s too short, medieval technology? How find mathematical formulas speed required to move a and the animations are can you build a pyramid out necessary to solve these submarine through enemy dated. But it was fun and well of massive stones using only problems as real engineers do. patrols on the first shot. It worth both the time to play sand, clay, and ropes? And Coby, my 7-year-old, instead took us about an hour and a it and the US $19.95 it costs. how do you inveigle kids, applied trial and error to the half to play the game through, Shingler says the money will 7 to 17, into tackling such basic engineering principles and we wanted more. go toward a flashier, longer questions? provided. He does have the Shingler says Time version due out next year. Game designer Ray engineering mind-set though, Engineers has been used by —Sherry Sontag www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 25 careers s s u US way, though, to Slim Devices, the tiny tiny the Devices, Slim to though, way, large a by over taken gets it when 100 Y handle just about any compression compression any about just handle with idea. Hecame hadup a different Calif., that created the Squeezebox, Squeezebox, the created that Calif., out, but it hasn’t sold out sold hasn’t it but out, music-playing the of instance, or designing one that could could that one designing or instance, suggesting began who programmers, up a storm, but selling already was iPod the company offered version$2000 a for the buck than anything else in the venture taking without product the which base, fan the of devotion the digital-to- a high-quality Slimp3, the for now-expanding market. For audiophiles, Sean Adams, a twentysomething geek, geek, a twentysomething Adams, Sean called Transporter in 2006. out and itin 2003, still gives more bang engineers. hiring and capital developing on keep to Adams enabled hit. a was too, it, and created, design— and code in changes and writers, magazine bloggers, of best-reviewed and first the of one View, Mountain in based company that happen didn’t It corporation. and started selling them online for for online them selling started and to homeconnection stereos. Ethernet an converter musicthat piped analog from decided to keep the gadget dumb. The dumb. The gadget to the keep decided customer even and testers, designers, streamers. audio digital-to-analog service reps. Having them around around them Having reps. service product of legion unpaid an as served Squeezebox the was Thus scheme. 26 lim Devices, maker maker Devices, lim queezebox, was bought bought was queezebox, nsqueezed Adams assembled the first first the assembled Adams The Slim Devices designers wisely wisely designers Devices Slim The The next Squeezebox, at came$299, Apple’s 2001. in story begins The

NA • iEEE Sp iEEE • NA One reason for the success was was success the for reason One

making a wireless version, for for version, wireless a making

$249. His creation caught the eye eye the caught creation His $249. units himself in his garage garage his in himself units lose its fans, crash, and burn burn and crash, fans, its lose ou might might ou maker of a niche gadget to to gadget niche a of maker E expect a garage-born garage-born a expect ctrum • • ctrum mA

y 2009 y

“a marketing guy, a sales guy, a support “Congrats, Slim Services employees,” Services Slim “Congrats, And because that software is written by written is software that because And They had 40 pages of something to say. of something pages 40 had They without turning on your computer. turning without to controller afreestanding adds which this hope only can “I Bklass. wrote MBAs, lawyers, and accountants get get accountants and lawyers, MBAs, lifetime. Personally, my heart sinks on sinks Personally, my heart lifetime. first 100 devices churned outin Sean’s regularly. be updated can models first posted someone with the handle Jasen. Jasen. handle the with someone posted and techies, of volunteer geeks, its base it was Sean and Dean Blackketter, an very for the even software the that ing the Squeezebox, letting you play music letting Squeezebox, the and critiqued, tested, recently base fan crowd is free-labor, this-is-our-gadget news.” this being rolled out, and the open-source, open-source, the out, and rolled being musicians. They had something to say. something had They musicians. now, Slim Devices wasn’t just Sean— company for $20 Squeezebox relies on your personal on your personal relies Squeezebox contained set of speakers, and the Duet, Duet, the and of speakers, set contained on the software himself, and is now early volunteer who bought one of the of larger payouts down the road. By on community, it keeps works and well music, mean the computer to process guy, and a testing guy,” Blackketter says. garage, learned Perl so he could work better. getting acquisition made your day, made acquisition year, change,” will things of Slim, ahold development open-source a passionate debugged the Boom, which has aself- has which Boom, the debugged director of engineering. There was also secret online forum of Slim Devices Devices of Slim forum online secret away. the blogging there, Indeed, still “Well, I’m saddened. Once the the Once “Well, I’m saddened. The Boom was vetted through a through vetted was Boom The But high-quality products are still still are products But high-quality by defined still was company But the InLogitech Octoberbought 2006, the

million, with promises - “Their day job is “Their that they get to do what “I with speakers. The secret forum was was forum secret The speakers. with with trepidation to whentrepidation with 2010, a worlds—way of toboth have best the Devices crew could crew receive as Devices much Maybe, butpain. some not without Bklass—whom to hired Logitech Pennsylvania, Texas, and Indianapolis. Pennsylvania, Texas, and Indianapolis. having fun here. Can’t speak for any for speak Can’t here. fun having not will he reiterated that Slim/Logitech official for many Logitech’s users—too Grumpy volunteers sometimes got into volunteers sometimes Grumpy posted how they’dloveposted aSqueezebox profits. Will theythetake moneyand profits. Will to own and pick apart a new cool toy. cool anew apart pick and to own from the technical discussions he discussions technical the from the message havefears filling been they did for free,” says Blackketter. telecommute placesfrom as far-flung the 10 former volunteers—including the but the process otherwise remained the the remained otherwise process but the board, enough awayto Adams board, pull new. So was a nondisclosure agreement, agreement, new. a nondisclosure was So of whom many already had regulars, run? Will it run? Will endmark the of Squeezebox one else.” one “sigh,”clearly With prefers. a written found themselves they as example, go closed-source. as open-source software? The fan base’s as open-sourceThe fan software? maker? electronics consumer a major the Unitedas Switzerland, Kingdom, as as $69.5 same—the fans worked for the chance worked chance fans for the same—the second payout Slim is due. The original to handle. channels support newbie more and more supporting for customers, paying with scrapes of resources corporate the and source

don’t even plan for next week. Still Still week. next for plan even don’t “As for anyone’s plans,” he wrote, wrote, he “Asplans,” anyone’s for The takeover was a windfall for The takeover was a windfall Meanwhile, volunteersthe look out a figured Squeezebox Has

passion and free labor of open of open labor free and passion

million more, depending onmore, depending million

www.spectrum.ieee.org —Sherry Sontag —Sherry -

logiteCh reflections By roBert w. lucky cloud computing

hey say that soon we The advantages for us, will do our computing the cloud’s clients, include T by renting memory the ability to add more and processing on a vast capacity for peak demand, cloud of computers out in the to flexibly experiment with network. They say this is an new services, and to remove idea whose time has come. unneeded capacity when Of course “they” have said demand slackens. Within “the time has come” about the giant cloud, the laws of other ideas, and some of those probability give the service things didn’t happen then, or provider great leverage sometimes ever. But this time through statistical multi- it’s for sure. They say. plexing of all these varying I remember the old days loads. The cloud is also easier of time-shared computing. to manage—you can install Big, expensive computers a single software patch to were kept behind big glass cover all of a company’s my own cycles and bytes; requires a subscription walls, tended by shrouded users, for example. I like being able to tend my service; the e-book reader acolytes. For the rest Those seem like own computing garden. whose crisp electronic of us, it was “keep your compelling arguments, Basically, the question is: page can display only hands off.” You rented so what could be wrong Do I trust these guys? books that come through computation by the second with this picture? For one Over the years, I’ve a particular vendor. All for your dumb terminal. thing, despite the existence always been both a system of these may be perfectly The miracle of Moore’s of encryption and access- designer and a system good systems, but they Law changed all that. control technologies, user, and I’ve seen that the don’t give me the flexibility Computing equipment has many organizations seem two sides are not always I want. So why should I become so cheap, the junk reluctant to put their congruent. The designer trust the cloud-computing piles are full of it. Who precious, proprietary data feels that the best system service providers? would have ever thought out in a public cloud. There is a closed one, where all There is a counter- of a terabyte of data just is a widely held impression the features and interfaces argument, of course. The lying around the house? that this might not be a are under his control. I cloud advocates say that So why should we go back good thing to be doing. don’t think that designers the open market will create to renting computational Also, there is a practical have evil intentions or even open systems; that openness power, this time from and administrative that they’re always trying will give some service a network cloud? problem—once all your to make more money for providers a competitive Apparently it’s precisely petabytes of data are out their companies; often the advantage. That may already because of Moore’s there in the cloud, can urge to control is simply be happening, as there are Law. Processors and you ever get them back? a part of the quest for different offerings that put memory are cheap these While corporate IT engineering perfection. the user at both the bottom days—and really cheap organizations ponder the No, it’s not the designers of the stack, where the when you buy, run, and advantages and drawbacks, I distrust so much as the access is effectively to a bare maintain them by the as an individual user I have a businesspeople. I’ve seen generic computer, and at the zillions. And the clouds different problem—giving up the cellphone that takes top, where the access is to at companies like Google my autonomy and control. pictures that can be sent customized applications.

john hersey john and Amazon are living I’ll admit from the start only to my computer, over So is cloud computing an things, constantly being that this is a psychological the contracted carrier, at idea whose time has come? replaced and upgraded hang-up; nevertheless, it’s a certain cost per picture; It’s your call. In the end, with the latest cheap stuff. there. I rather like having the built-in GPS that that’s the way it works. o

www.spectrum.ieee.org mAy 2009 • iEEE SPEctrum • NA 27 the million Dollar programming prize Netflix’s bounty for improving its movie- recommendation software is almost in the bag. Here is one team’s account

By RobeRt m. bell, JIm bennett, YehuDa KoRen & ChRIs VolInsKY

t’s 7:50 p.m. on 1 October 2007 at AT&T Labs, in Florham Park, N.J., and three of us are frantically hitting the “refresh” buttons on our browsers. We have just submitted our latest entry in the year-old Netflix Prize competition, which offers a grand prize of US $1 million for an algorithm that’s 10 percent more accurate than the one Netflix I uses to predict customers’ movie preferences. Although we have not reached that milestone, we are hoping at least to do better than anyone else has done so far; if we can make it to 8 p.m. with the best score, we’ll win a $50 000 Progress Prize. For most of the summer we’d been ahead of our nearest rivals by a comfortable margin, and as recently as 36 hours before this moment, our victory still seemed to be a slam dunk. The previous day, though, the lead had started to slip away from us. First, the teams then in fifth and sixth places merged, combining their tal- ents to vault into second place, making us nervous enough to submit our best effort, which we had been saving for a rainy day. But before our improved score appeared, we were hit by another combination when our two remain- ing serious rivals joined forces to tie us. Worse, their entry had come randi silberman

28 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 29 72 seconds before ours, meaning mender system, Cinematch, which will start recommending titles based that in the case of a tie, they’d win. helps customers get the most out of on the rating the algorithm predicts Seventy-two seconds! Could we their memberships. The better the the customer will give. lose this thing for being a min- system predicts people’s likes and ute too late? Then we realized that dislikes, the more they’ll enjoy the there were still 25 hours left, and movies they watch and the longer we still had one more chance. We they’ll keep up their subscriptions. ReCommenDIng moVIes that had to make it count. As a new subscriber to Netflix, customers will enjoy is a complex We began to receive offers from you have several ways to choose business. The easy part is gather- other competitors to combine our movies on the company’s Web ing the data, which Netflix now scores with theirs. We politely site. You can browse by genre accumulates at the rate of some declined them and planned strate- or search by keyword for a title, 2 million ratings a day. Much gies for our last submission. Sure actor, or director. After receiving tougher is to find patterns in the enough, these bumped up our your selections by mail or over the data that tell the company which score by a few hundredths of a Internet and viewing them, you movie a customer would enjoy, if percent, at which point we could return to the site to order more only the person would watch it. only wait to see the final score movies. At any time, Cinematch Netflix developed Cinematch to from our newly allied foes. lets you rate any movie you’ve seen tackle this job using a well-known Refresh...refresh...refresh... by clicking on one to five stars. prediction technique (described As is the case with other rec- below), which it designed to handle ommender systems, such as those the billion or so ratings it had already of Amazon, the Internet Movie logged. However, while incorporat- sInCe 1997, when netflIx, of Los Database, and Pandora, it’s in the ing so many ratings added to accu- Gatos, Calif., came up with the idea customer’s interest to vote thought- racy, it also took a lot of work just to of sending movie DVDs through fully, because doing so helps Netflix keep up with the increasing scale, the mail to subscribers, its cus- figure out his or her tastes. Yet even let alone to test alternative predic- tomer base has grown to 10 million. if the customer declines to offer this tion schemes. That success stems, in part, from feedback, Netflix still notes which The Netflix researchers were the company’s giving quick and movies the subscriber actually nevertheless curious about the easy access to movies. But just as orders. After the customer has rated many other techniques for making important is the Netflix recom- a handful of movies, the algorithm similar predictions that had been published in the scholarly literature. The problem was that those studies had relied on public data sets con- the neighborhood model taining on the order of a few mil- HE NEArESt-NEiGHBOr mEtHOD works on the principle that a person tends to lion ratings, and it would take the give similar ratings to similar movies. Joe likes the three movies on the left, so to small Netflix team a long time to make a prediction for him, find users who also liked those movies and see what t explore how well these alternative other movies they liked. Here the three other viewers all liked Saving Private Ryan, so that is the top recommendation. two of them liked Dune, so that’s ranked second, and so on. techniques worked at a scale a thou- photos: paramount pictures sand times as large. That is, if they did all the work themselves. Reed Hastings, the chief execu- tive of Netflix, suggested running a contest. He observed that Netflix had the means, the motive, and the opportunity. And the com- pany had already staged a contest internally between the standard Cinematch system and an alterna- tive algorithm, which did slightly, tantalizingly better. The Netflix team came up with the basic structure of the contest. They would provide 100 million rat- ings that 480 000 anonymous cus- tomers had given to 17 000 movies. The data set would just fit in the main memory of a typical laptop, allowing almost anyone to compete. Netflix would withhold 3 million of the most recent ratings and ask the contestants to predict them. A

30 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org Netflix computer would assess each the latent-factor approach contestant’s 3 million predictions by comparing predictions with actual SEcOND, cOmplEmENtAry method scores both a given movie and viewer according to latent factors, themselves inferred from the ratings given to all the ratings. The system would use the A movies by all the viewers. the factors define a space that at once measures the traditional metric for predictive characteristics of movies and the viewer’s interest in those characteristics. Here we accuracy, the root-mean squared would expect the fellow in the southeast corner of the graph to love Norbit, to hate error (RMSE). The more accurate Dreamgirls, and, perhaps, to rate Braveheart about average. photos: amadeus, the saul ZaentZ company; all others, paramount pictures a set of predictions, the smaller the

RMSE will be. The score would s e r i o u s then be reported back immediately to the contestant and reflected on an online leaderboard for all to see. Each such scoring provides valuable information about the hid- Amadeus den ratings—so valuable, in fact, that under certain circumstances Braveheart it could be used to game the system. Dreamgirls The teams were therefore scored once a day, at most. But to help teams estimate how well they might Mission do, Netflix also provided them each Impossible: 3 with a small representative data set and the score Cinematch had been s e i ov m y u g Terms of Endearment 48 Hours able to attain for it. Contestants could use that set to test their sys- tems as often as they wanted. chick flicks On 2 October 2006, Netflix launched the competition, and within days thousands of teams from hundreds of countries signed up. Within weeks the Netflix The Lion King Prize Web site was getting hun- dreds of submissions per day. An online forum was created so that participants could share ideas and techniques, even code. Even Norbit more gratifying to Netflix, within War of the Worlds months a handful of teams did sev- Crossroads eral percent better than Cinematch. The question then was how much e s c a p i s t the accuracy would improve in the first year. starring Tom Hanks. Its neigh- A second area of collaborative- bors may include other war movies, filtering research we pursued movies directed by Spielberg, or involves what are known as latent- movies starring Tom Hanks. To factor models. These score both lIKe most of the other top com- predict a particular viewer’s rat- a given movie and a given viewer petitors, the three of us at AT&T ing, we would look for the nearest according to a set of factors, them- Labs consulted the rich body of neighbors to Saving Private Ryan selves inferred from patterns in the research on ways of solving prob- that the viewer had already seen ratings given to all the movies by all lems in this domain, known as col- and rated. For some viewers, it the viewers [see illustration, “The laborative filtering. may be easy to find a full allotment Latent-Factor Approach”]. Factors One of the main areas of col- of close neighbors; for many oth- for movies may measure com- laborative filtering we exploited is ers, we may discover only a hand- edy versus drama, action versus the nearest-neighbor approach. A ful of neighboring movies. Our romance, and orientation to chil- movie’s “neighbors” in this context version of the nearest-neighbor dren versus orientation to adults. are other movies that tend to be approach predicted ratings using Because the factors are determined scored most similarly when rated a weighted average of the viewer’s automatically by algorithms, they by the same viewer [see illustra- previous ratings on up to 50 neigh- may correspond to hard-to-describe tion, “The Neighborhood Model”]. boring movies. (We have since concepts such as quirkiness, or they For example, consider Saving developed a way to use all past rat- may not be interpretable by humans Private Ryan (1998), a war movie ings, allowing an unlimited num- at all. Factors for viewers measure directed by Steven Spielberg and ber of neighbors.) how much the viewer likes movies www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 31 that score highly on the correspond- related films, such asThe Lord of the few well-selected strategies. ing movie factor. Thus, a movie may Rings trilogy (2001–2003). We don’t want to give the be classified a comedy, and a viewer Because these two methods impression that heaping together a may be classified as a comedy lover. are complementary, we combined lot of methods was enough to reach The model may use 20 to 40 them, using many versions of each the leaderboard. The Netflix Prize such factors to locate each movie in what machine-learning experts data set poses some huge chal- and viewer in a multidimensional call an ensemble approach. This lenges beyond its immense size. space. It then predicts a viewer’s allowed us to build systems that For one, there was enormous varia- rating of a movie according to the were simple and therefore easy to tion among viewers and among movie’s score on the dimensions code and fast to run. movies. A rating system must be that person cares about most. We What’s more, our ensemble sensitive enough to tease out sub- can put these judgments in quan- approach was robust enough to tle patterns associated with those titative terms by taking the dot protect against some of the prob- few viewers who rated 1500 movies, (or scalar) product of the locations lems that arise within the system’s without overfitting things—that is, of the viewer and the movie. individual components. Indeed, expecting prolific raters’ patterns Both collaborative-filtering tech- the solution we had just submit- to apply to the many more users niques work even if you don’t know ted on 1 October 2007 was a linear who rated 15 or fewer movies. It can a thing about what’s in the movies combination of 107 separate sets indeed be hard to make predictions themselves. All you need to care of predictions, using many varia- for a viewer who has provided just about is how the viewers rate the tions on the above themes and dif- a handful of ratings. We improved movies. However, neither approach ferent tuning parameters. Even existing ad hoc methods, designed is a panacea. We found that most so, the biggest improvements in to address this concern rigorously. nearest-neighbor techniques work accuracy came from relatively few Another critical innovation best on 50 or fewer neighbors, which methods. The lesson here is that involved focusing on which movies means these methods can’t exploit having lots of ways to skin this a viewer rated, regardless of the all the information a viewer’s rat- particular cat can be useful for scores. The idea is that some- ings may contain. Latent-factor gaining the incremental improve- one who has rated a lot of fantasy models have the opposite weakness: ments needed to win competitions, movies will probably like Lord of They are bad at detecting strong but practically speaking, excellent the Rings, even if that person has associations among a few closely systems can be built using just a rated the other movies in the cate-

a virtual “channel” by selecting a song, loves you,” by the same band. i listened for want artist, or composer. if a song is chosen, a long time before getting my first choice. the site compares it to its database you rate a song by clicking on either of 600 000 songs, each rated by one a thumbs-up or a thumbs-down icon, advice? of its musical experts. the site then and the algorithm adjusts its weighting selects another song it deems to be a of the musical checklist it uses to select try an expert close relative and keeps on playing such subsequent songs. What’s more, a relatives. (pandora can’t give you your first thumbs-down will keep the channel from f yOu want recommendations and choice because its licensing contracts ban ever playing the same song again. you you’d rather not rely exclusively on your it from playing songs to order.) have to be careful, because the more i customers, you can always do the daring When i selected “A Hard Day’s Night,” by thumbs-down you give, the narrower the thing and consult actual experts. that’s the Beatles, the first songi heard was “She channel becomes, and in the extreme case the idea behind pandora, a free internet you may “thumb yourself into a corner,” radio service that employs musicians to says tim Westergren, founder of pandora. rate songs according to a checklist of Even then, he notes, you’d only hobble criteria, such as pace, rhythm, even the that one channel. Nothing you do on one voice of the performer. channel affects the others, and you may the Oakland, calif.–based company create as many channels as you want. must be doing something right. in the Westergren says he got the idea for three years since it opened shop, pandora pandora when he was a young musician media has registered 22 million listeners. working on scores with moviemakers So far, all of them are in the united States, who had very different likes and dislikes. although the company is negotiating its He wanted to find a way to encode those way back into Europe, which it left after differences in a database he dubbed the having problems with music licenses there. music Genome, paying musicians to do About 2 million people listen to the the enormous amount of work. service on a given day, typically while it may seem strange to use so much sitting in front of their computers at work manpower as a supplement to computer or, increasingly, while clutching their power, but it makes sense when humans iphones on the commute home, making alone can handle the job—a peculiar for an average session of 6 hours. No wonder pandora streams more data than song central: tim Westergren any other site except youtube. and a few cDs waiting to be rated. Here’s how it works. the listener creates photo: rafael fuchs

32 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org gory somewhat low. By replacing continue to made progress. During have invigorated other fields. The numerical scores with a binary who- 2008 we mined the data for infor- various X Prizes that have been watched-what score, the data set is mation on how users’ behavior offered for advances in genomics, transformed from one with mostly changed over time. Later, we automotive design, and alterna- missing pieces (the cases in which joined forces with another team tive energy have shown an excel- users don’t rate movies) to one that to win the 2008 Progress Prize. lent return: By some accounts the is completely full (using the value 1 Currently we stand at 9.63 percent recent $10 million Ansari X prize, when there are ratings and a 0 when improvement and are still working awarded for suborbital spaceflight, there aren’t). This approach nicely hard on the problem. generated $100 million of private complemented our other methods. Now that the confetti has settled, investment in space travel. we have a chance to look back on The competition also validates our work and to ask what this expe- the concept of collective intelligence rience tells us. First, Netflix has popularized in James Surowiecki’s fInallY, at 7:58 on that fateful incorporated our discoveries into an 2005 book The Wisdom of Crowds October evening, all of the scores for improved version of its algorithm, (Anchor Books). He argues that the the top teams were posted for the last which is now being tested. Second, sum of many independent votes is time on the leaderboard, and ours researchers are benefiting from the often superior to any one vote, even came out highest, with an 8.43 per- data set that the competition made if made by the greatest expert. For cent improvement on Netflix’s algo- available, and not just because it is Netflix, the vast number of indepen- rithm. Our nearest rival scored an orders of magnitude larger than dent ratings allows for surprisingly 8.38 percent improvement. We previous data sets. It is also quali- good predictions. The power of this didn’t do well enough to land a mil- tatively better than other data sets, collective intelligence is also being lion dollars, but still, we won. because Netflix gathered the infor- harnessed in, for instance, Amazon. While we’ve since come very mation from paying customers, in a com’s product recommendations close to the goal of 10 percent, expe- realistic setting. Third, the compe- and the collaborative editing of the rience has shown that each step tition itself recruited many smart online encyclopedia, Wikipedia. forward is harder than the one that people in this line of research. With the rise of social networks on came before it, presumably because In any case, the new blood the Web, we can expect to see and we’ve already exploited most of the promises to quickly improve the contribute to even more powerful clues in the data. Nonetheless, we state of the art. Such competitions examples in the coming years. o

field sometimes called artificial artificial probably never get with other methods. intelligence. One example of AAi is setting He cites the ’80s pop star cindy lauper, puzzles, or “captchas,” for visitors to a Web who recently recorded a new record that site to solve, both to prove that they’re didn’t sell in great numbers. “We analyzed human beings and not bots and to perform it for its genome and found that the record some useful chore, such as deciphering the sounds an awful lot like Norah Jones. So blurred letters from a scan of an old book. we are able to play lauper’s songs when Other AAi programs lure people to do such you start a Norah Jones song. there’s a work by providing entertainment or, as metallica ballad that’s musically a nice Amazon’s mechanical turk does, money. fit for indigo Girls. So start an indigo Girls Westergren got seed money for the music station and you might get this ballad.” Genome in 2000, at the very end of the conrad says that pandora isn’t so dot-com bubble. When the bubble burst, he proud of its expert-rated system that it and his colleagues labored almost without can’t learn from the collaborative-filtering income for five years before another techniques pioneered at Amazon, Apple, injection of capital came through. Even now, music 101: pandora offers a running Netflix, and other firms. He contends that Westergren says, pandora is focused solely commentary on its songs and artists. the two approaches are complementary. on growth and so does not turn a profit.i t “We have benefited by peering inside the gets most of its revenue from the banner approaches tried by some of the thinking advertisements its site displays to listeners sell in very small numbers yet collectively that went into the Netflixp rize competition, every time they click on something in the amount to a big part of the online and we’ve incorporated some of the ideas site, something they must do from time to marketplace. mining that tail is one of the into our own system,” conrad says. “i’m time to prevent pandora from going silent. it main jobs of any recommender system. friendly with the Netflix personalization also gets a small royalty whenever a listener “in book publishing, genres are the team; we’ve talked over the past two buys a song by clicking through to a vendor, equivalent of what we’re doing. A brand- years or so. We wanted to have more such as itunes or Amazon. new author can say, ‘mine’s a historical qualitative information; they wanted more One advantage of using experts is that mystery novel,’ and thus put data into the quantitative. Now we both use both. Netflix they can categorize songs that are new, product without having any customer has human editors who try to capture by bands that are unknown. they can also reviews,” Westergren says. “But our theory technical aspects of the movies.” provide a way to get at music that fell out is that it’s not good data, not granular When the two approaches meet, of fashion before internet rating became enough and not objective.” experts will use computers as much as possible. Such too-new and too-old songs tom conrad, the chief technical officer computers use experts. We will have constitute a big part of the “long tail”— of pandora, says that “musical genomes” achieved the perfect chimera: a man- the huge inventory of items that each sometimes turn up connections you’d machine mind meld. —Philip E. Ross www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 33

hensible robot in “Futurama”? Answer: MOS Technology’s 6502. What these chips have in is that they’re part of the reason why engi- neers don’t get out enough. Of course, lists like this are nothing if not contentious. Some may accuse us of capricious choices and blatant omissions (and, no, it won’t be the first time). Why Intel’s 8088 microprocessor and not the 4004 (the first) or the 8080 (the famed)? Where’s the radiation-hardened military- grade RCA 1802 processor that was the brains of numerous spacecraft? If you take only one thing away from this introduction, let it be this: Our list is what remained after weeks of raucous debate between the author, his trusted sources, and several editors of IEEE Spectrum. We never intended to com- pile an exhaustive reckoning of every chip that was a commercial success or a major technical advance. Nor could we include chips that were great but so obscure that only the five engineers who Microchips designed them would remember them. We focused on chips that proved unique, intriguing, awe-inspiring. We wanted chips of varied types, from both big and small companies, created long ago or ThaT shook more recently. Above all, we sought ICs that had an impact on the lives of lots of people—chips that became part of earth- shaking gadgets, symbolized technologi- cal trends, or simply delighted people. For each chip, we describe how it came The World about and why it was innovative, with comments from the engineers and exec- utives who architected it. And because we’re not the IEEE Annals of the History n microchip design, as in life, groping for more technology clichés to of Computing, we didn’t order the 25 chips small things sometimes add up describe them. Suffice it to say that they chronologically or by type or importance; to big things. Dream up a clever gave us the technology that made our we arbitrarily scattered them on these microcircuit, get it sculpted in brief, otherwise tedious existence in this pages in a way we think makes for a good a sliver of silicon, and your lit- universe worth living. read. History is messy, after all. tle creation may unleash a tech- We’ve compiled here a list of 25 ICs As a bonus, we asked eminent tech- nological revolution. It happened with that we think deserve the best spot on nologists about their favorite chips. Ever the microprocessor. And the the mantelpiece of the house that Jack wonder which IC has a special place MK4096 4-kilobit DRAM. And Kilby and built. Some in the hearts of both , the TMS32010 digi- have become enduring objects of wor- of Intel, and Morris Chang, founder of tal signal processor. ship among the chiperati: the Signetics Taiwan Semiconductor Manufacturing Among the many great chips that 555 timer, for example. Others, such as Company? (Hint: It’s a DRAM chip.) have emerged from fabs during the half- the Fairchild 741 operational amplifier, We also want to know what you think. century reign of the , a became textbook design examples. Some, Is there a chip whose absence from our small group stands out. Their designs like Microchip Technology’s PIC micro- list sent you into paroxysms of rage? proved so cutting-edge, so out of the box, controllers, have sold billions, and are still Take a few deep breaths, have a nice so ahead of their time, that we are left doing so. A precious few, like Toshiba’s cup of chamomile tea, and then go to flash memory, created whole new mar- http://www.spectrum.ieee.org/may09/ kets. And one, at least, became a geeky 25chips. There you can drop us a line reference in popular culture. Question: and check out the runners-up that didn’t By Brian r. Santo What processor powers Bender, the alco- make the list and more favorite picks by holic, chain-smoking, morally repre- other luminaries. 34 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org Signetics NE555 Timer (1971)

It Was the summer of 1970, and chip designer Hans Camenzind could tell you a thing or two about Chinese restaurants: His small office was squeezed between two of them in downtown Sunnyvale, Calif. Camenzind was working as a consultant to Signetics, a local semi- conductor firm. The economy was At first, Signetics’s engineering tanking. He was making less than department rejected the idea. The US $15 000 a year and had a wife company was already selling com- and four children at home. He really ponents that customers could use to needed to invent something good. make timers. That could have been And so he did. One of the great- the end of it. But Camenzind insisted. est chips of all time, in fact. The 555 He went to Art Fury, Signetics mar- was a simple IC that could function keting manager. Fury liked it. as a timer or an oscillator. It would Camenzind spent nearly a year become a best seller in analog semi- testing breadboard prototypes, MOS Technology 6502 conductors, winding up in kitchen drawing the circuit components Microprocessor (1975) appliances, toys, spacecraft, and a on paper, and cutting sheets of few thousand other things. Rubylith—a masking film. “It was When the chubby-faced geek “And it almost didn’t get made,” all done by hand, no computer,” he stuck that chip on the computer and recalls Camenzind, who at 75 is still says. His final design had 23 tran- booted it up, the universe skipped a designing chips, albeit nowhere sistors, 16 resistors, and 2 diodes. beat. The geek was Steve Wozniak, the near a Chinese restaurant. When the 555 hit the market computer was the Apple I, and the chip The idea for the 555 came to him in 1971, it was a sensation. In 1975 was the 6502, an 8-bit microprocessor when he was working on a kind of Signetics was absorbed by Philips developed by MOS Technology. The system called a phase-locked loop. Semiconductors, now NXP, which chip went on to become the main brains With some modifications, the circuit says that many billions have been of ridiculously seminal computers like could work as a simple timer: You’d sold. Engineers still use the 555 to the Apple II, the Commodore PET, and trigger it and it would run for a cer- create useful electronic modules—as the BBC Micro, not to mention game sys- tain period. Simple as it may sound, well as less useful things like “Knight tems like the Nintendo and Atari. Chuck there was nothing like that around. Rider”–style lights for car grilles. Peddle, one of the chip’s creators, recalls when they introduced the 6502 at a trade show in 1975. “We had two glass jars learning toy. In the Steven Spielberg filled with chips,” he says, “and I had movie, the flat-headed alien uses it my wife sit there selling them.” Hordes to build his interplanetary commu- showed up. The reason: The 6502 wasn’t nicator. (For the record, E.T. also just faster than its competitors—it was

Clo uses a coat hanger, a coffee can, and also way cheaper, selling for US $25 C kwise from top left: Hans Camenzind; “ Camenzind; topHans left: from kwise a circular saw.) while Intel’s 8080 and Motorola’s 6800 The TMC0281 conveyed voice were both fetching nearly $200. f ox using a technique called linear pre- The breakthrough, says Bill Mensch, f ilm Corporation. dictive coding; the sound came out who created the 6502 with Peddle, was a as a combination of buzzing, hiss- minimal instruction set combined with a ing, and popping. It was a surpris- fabrication process that “yielded 10 times ing solution for something deemed as many good chips as the competition.” a ll “impossible to do in an integrated The 6502 almost single-handedly forced r ig

H circuit,” says Gene A. Frantz, one the price of processors to drop, helping ts f utur r of the four engineers who designed launch the personal computer revolu- eserved; eserved; the toy and is still at TI. Variants of tion. Some embedded systems still use a Texas Instruments TMC0281 ma” the chip were used in Atari arcade the chip. More interesting perhaps, the tm u Speech Synthesizer (1978) niversal/ and © 2009 games and Chrysler’s K-cars. In 6502 is the electronic brain of Bender, If It Weren’t for the TMC0281, 2001, TI sold its speech-synthesis the depraved robot in “Futurama,” as

tH E.T. would’ve never been able to chip line to Sensory, which dis- revealed in a 1999 episode. e e t wentiet k “phone home.” That’s because the continued it in late 2007. But if you [See “The Truth About Bender’s obal Colle obal TMC0281, the first single-chip ever need to place a long, very-long- Brain,” in this issue, where David X. H

Century speech synthesizer, was the heart distance phone call, you can find Cohen, the executive producer and head C tion (or should we say the mouth?) of Speak & Spell units in excellent con- writer for “Futurama,” explains how the

Texas Instruments’ Speak & Spell dition on eBay for about US $50. choice of the 6502 came about.] www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 35 Microchip Technology PIC 16C84 (1993)

back In the early 1990s, the huge 8-bit micro- controller universe belonged to one company, the almighty Motorola. Then along came a small contender with a non- descript name, Microchip Technology. Microchip devel- oped the PIC 16C84, which incorporated a type of mem- ory called EEPROM, for elec- trically erasable program- remote controls, and wireless mable read-only memory. It car keys. It was the beginning didn’t need UV light to be of a line of erased, as did its progenitor, that became electronics super- EPROM. “Now users could stars among Fortune 500 com- change their code on the fly,” panies and weekend hobbyists Texas Instruments TMS32010 says Rod Drake, the chip’s alike. Some 6 billion have been Digital Signal Processor (1983) lead designer and now a direc- sold, used in things like indus- tor at Microchip. Even better, trial controllers, unmanned the bIg state Of texas has given us many the chip cost less than US $5, aerial vehicles, digital preg- great things, including the 10-gallon hat, chicken- or a quarter the cost of exist- nancy tests, chip-controlled fried steak, Dr Pepper, and perhaps less prominently, ing alternatives, most of them fireworks, LED jewelry, and the TMS32010 digital signal processor chip. Created from, yes, Motorola. The 16C84 a septic-tank monitor named by Texas Instruments, the TMS32010 wasn’t the first found use in smart cards, the Turd Alert. DSP (that’d be Western Electric’s DSP-1, introduced in 1980), but it was surely the fastest. It could com- pute a multiply operation in 200 nanoseconds, a feat that made engineers all tingly. What’s more, it could μA741 Op-Amp (1968) execute instructions from both on-chip ROM and off-chip RAM, whereas competing chips had only OperatIOnal amplIfIers are the sliced bread of ana- canned DSP functions. “That made program devel- log design. You can always use some, and you can slap them opment [for the TMS32010] flexible, just like with together with almost anything and get something satisfying. microcontrollers and microprocessors,” says Wanda Designers use them to make audio and video preamplifiers, Gass, a member of the DSP design team, who is still voltage comparators, precision rectifiers, and many other sys- at TI. At US $500 apiece, the chip sold about 1000 tems that are part of everyday electronics. units the first year. Sales eventually ramped up, and In 1963, a 26-year-old engineer named Robert Widlar the DSP became part of modems, medical devices, designed the first monolithic op-amp IC, the μA702, at and military systems. Oh, and another application: Fairchild Semiconductor. It sold for US $300 a pop. Widlar Worlds of Wonder’s Julie, a Chucky-style creepy doll followed up with an improved design, the μA709, cutting the that could sing and talk (“Are we making too much cost to $70 and making the chip a huge commercial success. noise?”). The chip was the first in a large DSP family The story goes that the freewheeling Widlar asked for a raise. that made—and continues to make—TI’s fortune. When he didn’t get it, he quit. was ntersil i nology; nology; CH e

Intersil ICL8038 Waveform Generator beat the phone companies in the 1980s. t ip

(circa 1983*) The part was so popular the company CH ro C

put out a document titled “Everything i crItIcs scOffed at the ICL8038’s lim- You Always Wanted to Know About m aker;

ited performance and propensity for behaving the ICL8038.” Sample question: “Why b . erratically. The chip, a generator of sine, square, does connecting pin 7 to pin 8 give m triangular, sawtooth, and pulse waveforms, the best temperature performance?” eft: eft: Janet

was indeed a bit temperamental. But engineers Intersil discontinued the 8038 in 2002, l op

soon learned how to use the chip reliably, and but hobbyists still seek it today to make t the 8038 became a major hit, eventually selling things like homemade function genera- into the hundreds of millions and finding its tors and theremins. kwise from

way into countless applications—like the famed Moog music * Neither Intersil’s PR department nor the company’s last engineer C

synthesizers and the “blue boxes” that “phreakers” used to working with the part knows the precise introduction date. Do you? Clo 36 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org Western Digital Acorn Computers ARM1 WD1402A UART (1971) Processor (1985)

gOrdOn bell is famous for launch- In the early 1980s, Acorn ing the PDP series of minicomputers at Computers was a small company Digital Equipment Corp. in the 1960s. with a big product. The firm, based But he also invented a lesser known but in Cambridge, England, had sold no less significant piece of technology: over 1.5 million BBC Micro desk- the universal asynchronous receiver/ top computers. It was now time to transmitter, or UART. Bell needed some design a new model, and Acorn circuitry to connect a Teletype to a PDP-1, engineers decided to create their a task that required converting paral- own 32-bit microprocessor. They lel signals into serial signals and vice called it the Acorn RISC Machine, versa. His implementation used some or ARM. The engineers knew tion set, still remembers when they 50 discrete components. Western Digital, it wouldn’t be easy; in fact, they first tested the chip on a computer. a small company making chips, half expected they’d encounter an “We did ‘PRINT PI’ at the prompt, offered to create a single-chip UART. insurmountable design hurdle and it gave the right answer,” she Western Digital founder Al Phillips still and have to scrap the whole proj- says. “We cracked open the bottles remembers when his vice president of ect. “The team was so small that of champagne.” In 1990, Acorn spun engineering showed him the Rubylith every design decision had to favor off its ARM division, and the ARM sheets with the design, ready for fabri- simplicity—or we’d never finish it!” architecture went on to become the cation. “I looked at it for a minute and says codesigner Steve Furber, now dominant 32-bit embedded pro- spotted an open circuit,” Phillips says. a professor at cessor. More than 10 billion ARM “The VP got hysterical.” Western Digital the University of Manchester. In the cores have been used in all sorts of introduced the WD1402A around 1971, end, the simplicity made all the dif- gadgetry, including one of Apple’s and other versions soon followed. Now ference. The ARM was small, low most humiliating flops, the Newton UARTs are widely used in modems, PC power, and easy to program. Sophie handheld, and one of its most glit- peripherals, and other equipment. Wilson, who designed the instruc- tering successes, the iPhone.

only too happy to hire a guy who was then helping establish limits of semiconductor manufacturing processes at the the discipline of analog IC design. In 1967, Widlar created an time, incorporating a 30-picofarad capacitor into the chip. ever better op-amp for National, the LM101. Now, how to improve the front end? The solution was pro- While Fairchild managers fretted about the sudden com- foundly simple—“it just came to me, I don’t know, driving to petition, over at the company’s R&D lab a recent Tahoe”—and consisted of a couple of extra transis- hire, David Fullagar, scrutinized the LM101. He tors. That additional circuitry made the amplifica- realized that the chip, however brilliant, had a cou- tion smoother and consistent from chip to chip. ple of drawbacks. To avoid certain frequency distor- Fullagar took his design to the head of R&D at tions, engineers had to attach an external capacitor Fairchild, a guy named Gordon Moore, who sent it to the chip. What’s more, the IC’s input stage, the to the company’s commercial division. The new chip, so-called front end, was for some chips overly sen- the μA741, would become the standard for op-amps. sitive to noise, because of quality variations in the The IC—and variants created by Fairchild’s semiconductors. competitors—have sold in the hundreds of millions. Now, for “The front end looked kind of kludgy,” he says. $300—the price tag of that primordial 702 op-amp—you can get Fullagar embarked on his own design. He stretched the about a thousand of today’s 741 chips.

f could capture images at a resolution of

rom rom Kodak KAF-1300 1.3 megapixels, enough for sharp 5-by- t Image Sensor (1986) op: op: 7-inch prints. “At the time, 1 megapixel aC orn Computers; Computers; orn launched In 1991, the Kodak was a magic number,” says Eric Stevens, DCS 100 digital camera cost as much as the chip’s lead designer, who still works US $13 000 and required a 5-kilogram for Kodak. The chip—a true two-phase external data storage unit that users had charge-coupled device—became the d

avid to carry on a shoulder strap. The sight basis for future CCD sensors, helping to of a person lugging the contraption? Not kick-start the digital photography revo- f ullagar; ullagar; a Kodak moment. Still, the camera’s lution. What, by the way, was the very electronics—housed inside a Nikon F3 first photo made with the KAF-1300? k odak body—included one impressive piece “Uh,” says Stevens, “we just pointed the of hardware: a thumbnail-size chip that sensor at the wall of the laboratory.” www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 37 My Favorite

Gordon Moore cofounder Carver Mead and chairman emeritus of Intel professor of engineering and applied science, Caltech There were lots of great chips, but one that will always be dear to me was the , the My favorite chip contained only first commercial 1024-bit dram [introduced a single transistor, although in 1970]. It was the chip that really got Intel over a remarkable one: a shottky the hump to profitability. It was not the most ele- barrier gate field-effect gant design, having many of the problems that transistor made from GaAs memory engineers had become familiar with in (later called the MESFET and core memories. But this was comforting to such now, using more advanced engineers—it meant that their expertise was not semiconductor structures, going to be made obsolete by the new technology. Even today, when I look at the HEMT). I designed it over my digital watch and see 11:03, I cannot help but remember this key product Thanksgiving break in 1965. The very high mobility in Intel’s history. of the III-IV materials, together with the absence of minority-carrier storage effects, made these devices far superior for a microwave power-output stage. Despite my efforts to interest American companies, Lee Felsenstein the Japanese were the first to develop these devices. computer pioneer They have made microwave communications in sat- ellites, cellphones, and many other systems possible The humble, ubiquitous, and for many decades. quite inexpensive signetics 555 timer had a big impact on my career, when I found myself using it to craft var- Sophie Vandebroek ious pulse-sequencing cir- chief technology officer, cuits, baud-rate oscillators, and ramp generators for my Xerox

My selection is analog Jeff Hawkins devices’ imems accel- founder of Palm erometer, the first com- and Numenta mercial chip to significantly integrate MEMS and logic My personal favorite—a chip circuitry. Commercialized in that opened my eyes to what the early 1990s, it revolution- was possible—was the Intel ized the automotive air-bag 2716 eprOm, vintage late industry—and saved lives! Today this type of accel- earliest consulting clients. 1970s. The 2716 was nonvola- erometer is used in a variety of applications, includ- With a latch triggered by two tile, held 2 kilobytes of mem- ing the Nintendo Wii and the Apple iPhone. Other comparators, a high-current- ory, and, unlike many other types of MEMS chips are more and more present in drive output and a separate chips, used a single 5-volt a broad variety of applications. open-collector output for power supply, so it could be discharging capacitors, the used as storage in a small 555 offered a wide range of and practical package. Of uses with the addition of a course, its big disadvantage Steve Jurvetson few analog components. It was that you needed a UV managing director of suggested that it could pro- light to erase it. But with a vide more than it delivered little imagination you could Draper Fisher Jurvetson

at times, which helped me see that one day it could all erox x refine my understanding of be done electrically. In some The motorola 68000 was a the limits of a chip and the sense the 2716 is the great- special one. I built a speaking entures; umenta; umenta;

computer with it. I also wrote v

need to consider its operation grandparent of today’s flash n in the analog domain. memory chips. a multitasker in assembler. That chip was the workhorse osla kH ; of learning back during my J elsenstein; f M.S.E.E. days. But wait— mpin i ee

other chips come to mind. l

founder of TSMC C; Morris Chang ntel; I also like ZettaCore’s first i tsm

1-megabyte molecular memory chip, D-Wave’s solid- eft: l

One of my favorite great microchips is the Intel 1103, le;

state 128-qubit quantum computer processor, the o op

1-kilobit dram, circa 1970. Reasons: one, huge com- t (I’ve got an 8-inch wafer signed by Andy eino

mercial success; two, started Intel on its way; three, l Grove in my office), the Canon EOS 5D 12.8-megapixel demonstrated the power of MOS technology (versus CMOS sensor, and the iTV 5-bit asynchronous pro- itzel;

bipolar); and four, opened up at least another 40 years d cessor running Forth and a custom OS—whew! kwise from of life for Moore’s Law. C avid avid Clo d 38 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org IBM Deep Blue 2 Chess Chip (1997)

On One sIde of the board, 1.5 kilograms of gray matter. On the other side, 480 chess chips. Humans finally fell to computers in 1997, when IBM’s chess-playing com- puter, Deep Blue, beat the reigning world Vinod Khosla champion, Garry Kasparov. Each of cofounder of Sun Deep Blue’s chips consisted of 1.5 million Microsystems and arranged into specialized partner at Kleiner, Perkins, blocks—like a move-generator logic Caufield & Byers array—as well as some RAM and ROM. Deep Blue decisive—Kasparov called Together, the chips could churn through them “uncomputerlike”—moves. “They 200 million chess positions per second. exerted great psychological pressures,” That brute-force power, combined with recalls Deep Blue’s mastermind, Feng- clever game-evaluation functions, gave hsiung Hsu, now at Microsoft.

Texas Instruments Digital Micromirror The motorola 68010 was my favorite chip. With its Device (1987) power and virtual mem- On 18 June 1999, Larry ory support, it said to the world, “Microprocessors Hornbeck took his wife, Laura, can stand and compete with on a date. They went to watch the big boys—the minis and Star Wars: Episode 1—The mainframes.” Phantom Menace at a theater in Burbank, Calif. Not that the graying engineer was an avid Jedi fan. The reason they were there was actually the projec- tor. It used a chip—the digi- tal micromirror device—that Hornbeck had invented at Texas Instruments. The chip Transmeta Corp. uses millions of hinged micro- Crusoe Processor (2000) scopic mirrors to direct light through a projection lens. The David Ditzel WIth great pOWer come great screening was “the first digital Intel’s chief architect for heat sinks. And short battery life. And exhibition of a major motion hybrid parallel computing crazy electricity consumption. Hence picture,” says Hornbeck, a TI Transmeta’s goal of designing a low- Fellow. Now movie projectors My favorite chip: the 6502 power processor that’d put those hogs using this digital light-process- from mOs technology. offered by Intel and AMD to shame. ing technology—or DLP, as TI This 8-bit microproces- sor was used by me and The plan: Software would translate branded it—are used in thou- many hobbyists for build- instructions on the fly into Crusoe’s own sands of theaters. It’s also used f ing our own computers rom machine code, whose higher level of paral- in rear-projection TVs, office

[above, me and my home- t op: op: lelism would save time and power. It was projectors, and tiny projectors made PC in 1977]. I wrote a a

dam dam hyped as the greatest thing since sliced sil- for cellphones. “To para- small operating system that icon, and for a while, it was. “Engineering phrase Houdini,” Hornbeck fit in 4 kilobytes, and Tiny n adel/ BASIC from Tom Pittman Wizards Conjure Up Processor Gold” was says, “micromirrors, gen-

fit in 2 KB. The reason it ap how IEEE Spectrum’s May 2000 cover tlemen. The effect is created

was a great chip was that in pH oto; put it. Crusoe and its successor, Efficeon, with micromirrors.” the 8-bit era, you could use the first 256 bytes of mem- ieee “proved that dynamic binary translation

s was commercially viable,” says David

ory as 128 16-bit pointers pe C

to index from, making this trum; Ditzel, Transmeta’s cofounder, now at Intel. machine much easier to pro- Unfortunately, he adds, the chips arrived

gram than other 8-bit alter- t exas several years before the market for low- natives. It was a big deal i when you had to enter your nstruments power computers took off. In the end, while programs in hexadecimal Transmeta did not deliver on its prom- binary form. ises, it did force Intel and AMD—through licenses and lawsuits—to chill out. www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 39 Micronas Semiconductor MAS3507 MP3 Decoder (1997)

befOre the ipOd, there was the Diamond Rio PMP300. Not that you’d remember. Launched in 1998, the PMP300 became an instant hit, but then the hype faded faster than Milli Vanilli. One thing, though, was notable about the player. It carried the MAS3507 MP3 decoder chip—a RISC-based digital signal pro- cessor with an instruc- tion set optimized for audio compression and decompression. The chip, developed by Micronas, let the Rio squeeze about a dozen songs onto its flash memory—laughable today but at the time just Intel 8088 enough to compete with portable CD players. Quaint, huh? The Microprocessor (1979) Rio and its successors paved the way for the iPod, and now you can carry thousands of songs—and all of Milli Vanilli’s albums Was there any one chip that propelled Intel and music videos—in your pocket. into the Fortune 500? Intel says there was: the 8088. This was the 16-bit CPU that IBM chose for its original PC line, which went on to dominate Mostek MK4096 4-Kilobit DRAM (1973) the desktop computer market. In an odd twist of fate, the chip that established mOstek Wasn’t the fIrst to put out a DRAM. Intel what would become known as the x86 architec- was. But Mostek’s 4-kilobit DRAM chip brought about a key ture didn’t have a name appended with an “86.” innovation, a circuitry trick called address multiplexing, con- The 8088 was basically just a slightly modified cocted by Mostek cofounder Bob 8086, Intel’s first 16-bit CPU. Or as Intel engineer Proebsting. Basically, the chip Stephen Morse once put it, the 8088 was “a cas- used the same pins to access the trated version of the 8086.” That’s because the memory’s rows and columns new chip’s main innovation wasn’t exactly a step by multiplexing the address- forward in technical terms: The 8088 processed ing signals. As a result, the chip data in 16-bit words, but it used an 8-bit exter- wouldn’t require more pins as

nal data bus. memory density increased and ostek m

Intel managers kept the 8088 project under could be made for less money. om; wraps until the 8086 design was mostly com- There was just a little compati- C plete. “Management didn’t want to delay the 8086 bility problem. The 4096 used

by even a day by even telling us they had the 8088 16 pins, whereas the memories eim/eirikso. H ol

variant in mind,” says Peter A. Stoll, a lead engi- made by Texas Instruments, s

neer for the 8086 project who did some work on Intel, and Motorola had 22 pins. irik e the 8088—a “one-day task force to fix a microcode What followed was one of the most epic face-offs in DRAM history. bug that took three days.” With Mostek betting its future on the chip, its executives set out to ntel Corp.; It was only after the first functional 8086 came proselytize customers, partners, the press, and even its staff. Fred i eft: eft:

out that Intel shipped the 8086 artwork and docu- K. Beckhusen, who as a recent hire was drafted to test the 4096 l op

mentation to a design unit in Haifa, Israel, where devices, recalls when Proebsting and chief executive L.J. Sevin t two engineers, Rafi Retter and Dany Star, altered came to his night shift to give a seminar—at 2 a.m. “They boldly the chip to an 8-bit bus. predicted that in six months no one would hear or care about

The modification proved to be one of Intel’s 22-pin DRAM,” Beckhusen says. They were right. The 4096 and kwise from C

best decisions. The 29 000-transistor 8088 CPU its successors became the dominant DRAM for years. Clo required fewer, less expensive support chips than the 8086 and had “full compatibility with 8-bit hardware, while also providing faster processing and a smooth transition to 16-bit processors,” as Xilinx XC2064 FPGA (1985) Intel’s Robert Noyce and Ted Hoff wrote in a 1981 article for IEEE Micro magazine. back In the early came up with a chip packed The first PC to use the 8088 was IBM’s Model 1980s, chip designers tried to with transistors that formed 5150, a monochrome machine that cost US $3000. get the most out of each and loosely organized logic blocks Now almost all the world’s PCs are built around every transistor on their cir- that in turn could be config- CPUs that can claim the 8088 as an ancestor. Not cuits. But then Ross Freeman ured and reconfigured with bad for a castrated chip. had a pretty radical idea. He software. Sometimes a bunch 40 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org Z80 , another Microprocessor (1976) ex-Intel engineer, worked 80-hour weeks hunched over knew tables, drawing the Z80 cir- well the kind of money and cuits. Faggin soon learned man-hours it took to market that when it comes to micro- a microprocessor. While at chips, small is beautiful—but Intel, he had contributed to it can hurt your eyes. Sun Microsystems SPARC Processor (1987) the designs of two seminal “By the end I had to get there Was a tIme, long ago (the early 1980s), specimens: the primordial glasses,” he says. “I became when people wore neon-colored leg warmers and 4004, and the 8080, of Altair nearsighted.” watched “Dallas,” and microprocessor architects sought fame. So when he founded The team toiled through to increase the complexity of CPU instructions as a way Zilog with former Intel col- 1975 and into 1976. In March of getting more accomplished in each compute cycle. But league Ralph Ungermann, of that year, they finally then a group at the University of California, Berkeley, they decided to start with had a prototype chip. The always a bastion of counterculture, called for the opposite: something simpler: a single- Z80 was a contemporary of Simplify the instruction set, they said, and you’ll process chip microcontroller. MOS Technology’s 6502, and instructions at a rate so fast you’ll more than compen- Faggin and Ungermann like that chip, it stood out not sate for doing less each cycle. The Berkeley group, led by David Patterson, called their approach RISC, for reduced- instruction-set computing. As an academic study, RISC sounded great. But was it marketable? Sun Microsystems bet on it. In 1984, a small team of Sun engineers set out to develop a 32-bit RISC processor called SPARC (for Scalable Processor Architecture). The idea was to use the chips in a new line of workstations. One day, Scott McNealy, then Sun’s rented an office in downtown only for its elegant design CEO, showed up at the SPARC development lab. “He Los Altos, Calif., drafted a busi- but also for being dirt cheap said that SPARC would take Sun from a $500-million-a- ness plan, and went looking for (about US $25). Still, getting year company to a billion-dollar-a-year company,” recalls venture capital. They ate lunch the product out the door took Patterson, a consultant to the SPARC project. at a nearby Safeway super- a lot of convincing. “It was If that weren’t pressure enough, many outside Sun market—“Camembert cheese just an intense time,” says had expressed doubt the company could pull it off. Worse and crackers,” he recalls. Faggin, who developed an still, Sun’s marketing team had had a terrifying realiza- But the engineers soon ulcer as well. tion: SPARC spelled backward was…CRAPS! Team realized that the microcon- But the sales eventu- members had to swear they would not utter that word troller market was crowded ally came through. The to anyone even inside Sun—lest the word get out to arch- with very good chips. Even Z80 ended up in thousands rival MIPS Technologies, which was also exploring the if theirs was better than the of products, including the RISC concept. others, they’d see only slim Osborne I (the first por- The first version of the minimalist SPARC consisted of profits—and continue lunch- table, or “luggable,” com- a “20 000-gate-array processor without even integer mul-

left: left: ing on cheese and crackers. puter), and the Radio Shack tiply/divide instructions,” says Robert Garner, the lead

C Zilog had to aim higher on TRS-80 and MSX home com- SPARC architect and now an IBM researcher. Yet, at 10 pu-world. the food chain, so to speak, puters, as well as printers, million instructions per second, it ran about three times and the Z80 microprocessor fax machines, photocopiers, as fast as the complex-instruction-set computer (CISC) C

om; rig om; project was born. modems, and satellites. Zilog processors of the day. The goal was to outper- still makes the Z80, which is Sun would use SPARC to power profitable work- H t: form the 8080 and also offer popular in some embedded stations and servers for years to come. The first SPARC- r obert obert full compatibility with 8080 systems. In a basic configu- based product, introduced in 1987, was the Sun-4 line of software, to lure customers ration today it costs $5.73— workstations, which quickly dominated the market and g arner away from Intel. For months, not even as much as a cheese- helped propel the company’s revenues past the billion- Faggin, Ungermann, and and-crackers lunch. dollar mark—just as McNealy had prophesied.

of transistors wouldn’t be Xilinx. (Apparently, a weird using XC2064’s logic blocks, was only too happy to help used—heresy!—but Freeman concept called for a weird just as Xilinx customers would. the boss. “There we were,” he was betting that Moore’s Law company name.) When the Bill Carter, a former chief tech- says, “with paper and colored would eventually make tran- company’s first product, the nology officer, recalls being pencils, working on Bernie’s sistors really cheap. It did. To XC2064, came out in 1985, approached by CEO Bernie assignment!” Today FPGAs— market the chip, called a field- employees were given an Vonderschmitt, who said he sold by Xilinx and others—are programmable gate array, or assignment: They had to draw was having “a little difficulty used in just too many things to FPGA, Freeman cofounded by hand an example circuit doing his homework.” Carter list here. Go reconfigure! www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 41 Amati Communications Overture ADSL Chip Set (1994)

remember When DSL came along and you chucked that pathetic 56.6-kilobit-per-second modem into the trash? You and the two-thirds of the world’s broadband users who use DSL should thank Amati Communications, a start-up out of Stanford University. In the 1990s, it came up with a DSL modulation approach called discrete multi- Tripath Technology tone, or DMT. It’s basically a way of making one phone line look like hundreds of subchannels and improving TA2020 Audio transmission using an inverse Robin Hood strategy. “Bits Amplifier (1998) are robbed from the poorest channels and given to the wealthiest channels,” says John M. Cioffi, a cofounder of Amati and now an engi- there’s a subset of neering professor at Stanford. DMT beat competing approaches—including ones from audiophiles who insist that giants like AT&T—and became a global standard for DSL. In the mid-1990s, Amati’s vacuum tube–based ampli- DSL chip set (one analog, two digital) sold in modest quantities, but by 2000, volume fiers produce the best sound had increased to millions. In the early 2000s, sales exceeded 100 million chips per and always will. So when year. Texas Instruments bought Amati in 1997. some in the audio commu- nity claimed that a solid-state class-D amp concocted by a company called Tripath Technology delivered sound as warm and vibrant as tube amps, it was a big deal. Tripath’s trick was to use a 50-megahertz sampling Motorola MC68000 Microprocessor (1979)

system to drive the amplifier. atnagar The company boasted that mOtOrOla was late to the 16-bit micro- find a credit-card-size copy of the flow- bH avi avi its TA2020 performed better processor party, so it decided to arrive in charts sitting on my desk,” Tredennick r and cost much less than any style. The hybrid 16-bit/32-bit MC68000 recalls. The 68000 found its way into all useum; comparable solid-state amp. packed in 68 000 transistors, more than the early Macintosh computers, as well m To show off the chip at trade double the number of Intel’s 8086. It had as the Amiga and the Atari ST. Big sales shows, “we’d play that song— internal 32-bit registers, but a 32-bit bus numbers came from embedded applica- that very romantic one from would have made it prohibitively expen- tions in laser printers, arcade games, and Titanic,” says Adya Tripathi, sive, so the 68000 used 24-bit address industrial controllers. But the 68000 was

Tripath’s founder. Like most and 16-bit data lines. The 68000 seems also the subject of one of history’s great- Computerow; History H class-D amps, the 2020 was to have been the last major processor est near misses, right up there with Pete eter C very power efficient; it didn’t designed using pencil and paper. “I cir- Best losing his spot as a drummer for the p ; require a heat sink and culated reduced-size copies of flowcharts, Beatles. IBM wanted to use the 68000 in H ripat

could use a compact package. execution-unit resources, decoders, and its PC line, but the company went with t Tripath’s low-end, 15-watt control logic to other project members,” Intel’s 8088 because, among other things, version of the TA2020 sold for says Nick Tredennick, who designed the 68000 was still relatively scarce. US $3 and was used in boom the 68000’s logic. The copies were small As one observer later reflected, had boxes and ministereos. Other and difficult to read, and his bleary-eyed Motorola prevailed, the Windows-Intel kwise fromleft: top

versions—the most powerful colleagues found a way to make that duopoly known as Wintel might have C

had a 1000-W output—were clear. “One day I came into my office to been Winola instead. Clo used in home theaters, high- end audio systems, and TV sets by Sony, Sharp, Toshiba, and others. Eventually, the big semiconductor compa- Chips & Technologies AT Chip Set (1985) nies caught up, creating simi- lar chips and sending Tripath by 1984, when IBM introduced its 80286 AT line of PCs, into oblivion. Its chips, how- the company was already emerging as the clear winner in ever, developed a devoted desktop computers—and it intended to maintain its domi- cult following. Audio-amp nance. But Big Blue’s plans were foiled by a tiny company kits and products based on called Chips & Technologies, in San Jose, Calif. C&T devel- the TA2020 are still available oped five chips that duplicated the functionality of the AT from such companies as 41 Hz motherboard, which used some 100 chips. To make sure Audio, Sure Electronics, and the chip set was compatible with the IBM PC, the C&T Winsome Labs. engineers figured there was just one thing to do. “We had 42 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org Computer Cowboys Sh-Boom Toshiba NAND Flash Memory (1989) Processor (1988) the saga that is the invention into production and tWO chIp desIgners walk into a bar. They invention of flash memory watched as the money poured are Russell H. Fish III and Chuck H. Moore, and the began when a Toshiba fac- in, you don’t know much bar is called Sh-Boom. No, this is not the beginning of tory manager named Fujio about how huge corporations a joke. It’s actually part of a technology tale filled with Masuoka decided typically exploit discord and lawsuits, lots of lawsuits. It all started in he’d reinvent semi- internal innova- 1988 when Fish and Moore created a bizarre proces- conductor memory. tions. As it turned sor called Sh-Boom. The chip was so streamlined it We’ll get to that in a out, Masuoka’s could run faster than the clock on the circuit board minute. First, a bit bosses at Toshiba that drove the rest of the computer. So the two design- (groan) of history is told him to, well, ers found a way to have the processor run its own in order. erase the idea. superfast internal clock while still staying synchro- B e fo r e f l a s h He didn’t, of nized with the rest of the computer. Sh-Boom was m e m o r y c a m e course. In 1984 never a commercial success, and after patenting its along, the only way to store he presented a paper on his innovative parts, Moore and Fish moved on. Fish what then passed for large memory design at the IEEE later sold his patent amounts of data was to use International Electron Devices rights to a Carlsbad, magnetic tapes, floppy disks, Meeting, in San Francisco. Calif.–based firm, and hard disks. Many com- That prompted Intel to begin Patriot Scientific, panies were trying to create development of a type of flash which remained a solid-state alternatives, but memory based on NOR logic profitless speck of the choices, such as EPROM gates. In 1988, the company a company until its (or erasable programmable introduced a 256-kilobit chip executives had a rev- read-only memory, which that found use in vehicles, elation: In the years required ultraviolet light to computers, and other mass- since Sh-Boom’s erase the data) and EEPROM market items, creating a nice invention, the speed (the extra E stands for “elec- new business for Intel. of processors had by trically,” doing away with That’s all it took for far surpassed that of the UV) couldn’t store much Toshiba to finally decide to motherboards, and data economically. market Masuoka’s invention. so practically every Enter Masuoka-san at His flash chip was based on maker of computers Toshiba. In 1980, he recruited NAND technology, which and consumer elec- four engineers to a semisecret offered greater storage den- tronics wound up project aimed at designing sities but proved trickier to using a solution just like the one Fish and Moore had a memory chip that could manufacture. Success came patented. Ka-ching! Patriot fired a barrage of lawsuits store lots of data and would in 1989, when Toshiba’s first left: C left: against U.S. and Japanese companies. Whether these be affordable. Their strategy NAND flash hit the market.

H companies’ chips depend on the Sh-Boom ideas is a was simple. “We knew the cost And just as Masuoka had pre- u C k k matter of controversy. But since 2006, Patriot and of the chip would keep going dicted, prices kept falling. m oore; rig oore; Moore have reaped over US $125 million in licensing down as long as transistors Digital photography gave fees from Intel, AMD, Sony, Olympus, and others. As shrank in size,” says Masuoka, flash a big boost in the late

H for the name Sh-Boom, Moore, now at IntellaSys, in now CTO of Unisantis 1990s, and Toshiba became t: f u Cupertino, Calif., says: “It supposedly derived from Electronics, in Tokyo. one of the biggest players in J io masuoka masuoka io the name of a bar where Fish and I drank bourbon Masuoka’s team came up a multibillion-dollar mar- and scribbled on napkins. There’s little truth in that. with a variation of EEPROM ket. At the same time, though, But I did like the name he suggested.” that featured a Masuoka’s relationship with consisting of a single transis- other executives soured, and tor. At the time, conventional he quit Toshiba. (He later sued EEPROM needed two tran- for a share of the vast profits sistors per cell. It was a seem- and won a cash payment.) ingly small difference that had Now NAND f lash is a a huge impact on cost. key piece of every gadget— the nerve-racking but admittedly enter- In search of a catchy name, cellphones, cameras, music taining task of playing games for weeks,” they settled on “flash” because players, and of course, the USB says Ravi Bhatnagar, the chip-set lead of the chip’s ultrafast eras- drives that techies love to wear designer and now a vice president at ing capability. Now, if you’re around their necks. “Mine has Altierre Corp., in San Jose, Calif. The thinking Toshiba rushed the 4 gigabytes,” Masuoka says. o C&T chips enabled manufacturers like Taiwan’s Acer to make cheaper PCs and launch the invasion of the PC clones. Intel With additional reporting by Sally Adee, Erico Guizzo, and Samuel K. Moore bought C&T in 1997. www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 43 44 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org N e x t- G e N

U l t r a s o UN d

medical imaging borrows lmoSt iNvAriAbly, a new baby’s photo album begins with techniques from the a grainy black-and-white picture taken months before birth— A a prenatal ultrasound image, which is often detailed enough to microelectronics industry inspire comments about the child’s resemblance to various members of By Butrus t. Khuri-yaKuB, the family. But jokes about balding uncles notwithstanding, such scans Ömer OralKan & mariO KupniK serve a serious purpose and can prove immensely important, as when they allow doctors to diagnose and sometimes even repair a congenital malformation while the baby is still in the womb. When seeing such an image for the first time, most people are awe- struck. How can mere sound waves provide such remarkably clear views? Engineers may well ask something more: How can we give doc- tors even better ultrasound images? That question has engaged the three of us, along with other members of our Stanford acoustics group, for SHARPER much of the last decade. IMAGE: Whereas the signal-processing and image-reconstruction tech- micromachined niques used in medical ultrasonography have made huge advances transducer probes for since this type of imaging became commonplace three decades ago, the ultrasound business end of the apparatus—the transducer, which converts elec- scanners should trical impulses to sound waves and vice versa—has remained largely provide prenatal unchanged. So we found fertile ground when we began digging for ways images that are to improve those transducers using tools from the microelectronics even sharper than those new industry. You will soon find the fruits of those efforts at your local hos- parents now pital. Indeed, this strategy promises to revolutionize ultrasound imag- get to see. the ing within the next few years. pictures, though, may never be as crisp as ow ultrASouNd imaging works is easy enough to describe, at the one in this H least in broad strokes. High-frequency (1- to 50-megahertz) sound fanciful photo- waves transmitted into the body create reflections when they encoun- illustration. ter a change in tissue density or stiffness. These faint echoes are picked Photo-IllustratIon: Paul VozdIc/Getty up with the same set of transducers used to generate the sound. Or the ImaGes www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 45 Catheter Forward- A CATHETER’S looking ulTRASonIC EyES transducer micromachined capacitive transducers array can be easily formed into arrays of various shapes and sizes. Small ring- shaped arrays, for example, can provide forward-looking ultrasonic views for doctors probing an artery with a catheter. dangerous blockages and certain kinds of heart problems can then be treated using surgical tools that pass through the center of the catheter. the elements of the array depicted Surgical in this drawing [blue rectangles] are Blood tool each composed of many individual vessel transducer units [small brown squares in the inset photograph].

imager may use just a single transducer moved over single plane to make a two-dimensional image, a 2-D the body—usually with the aid of much slimy goo, to array can be steered and focused throughout a volume ensure good acoustic coupling. The resulting elec- to make a three-dimensional image—and this can be trical signals are then amplified, combined, and dis- done in real time. played as images. With this capability, physicians can, for example, Ultrasonography is valuable for several reasons. follow heart motions in great detail if they want to For one, it’s inexpensive—at least compared with CT assess a patient’s cardiac functioning. In the not-so- (computed tomography) and PET (positron-emission distant future, such ultrasound imaging may even tomography) scanning, or with MRI (magnetic reso- allow robotic surgeons to operate on a beating heart nance imaging). Also, the low-amplitude ultrasound so that patients need not run the risk of having to waves used for imaging do not involve ionizing radia- depend on a heart-lung machine. tion and are thus harmless to the patient, so repeated In the nearer term, doctors are keen to use small scans can be made without worry. And with this 2-D arrays of tiny ultrasonic transducers to obtain technique it is not difficult to get real-time imagery, forward-looking images as they probe an artery which doctors may want for such things as guiding with a catheter. That would permit them to exam- a biopsy needle. These virtues make the market for ine obstructions and map the composition of plaque medical ultrasound equipment huge—more than deposits on vessel walls in three dimensions. What’s US $5 billion annually, a figure that’s only expected more, sufficiently small transducers can be arranged to swell in coming years with growing sales of these in a ring on the end of a catheter, leaving space at systems in China and India. the center for an excision device. Such an instrument An ultrasound imager has four main parts: the would allow for simultaneous ultrasound imaging transducer probe, the analog front-end electronics, and surgical therapy. the digital signal-processing hardware, and the dis- Two-dimensional arrays of ultrasonic transducers play. Advances in electronics over the years have would certainly help physicians perform minimally brought an extraordinary level of refinement to all invasive treatments in this way. But making such tiny but the transducer, which means that most of the arrays using traditional transducers is frustratingly remaining opportunities for improving system per- difficult. Fortunately, the precise fabrication required formance lie in the design of this one critical com- can readily be carried out using methods developed ponent. In particular, researchers have lately been by the microelectronics industry, methods that are seeking reliable ways to fashion many individual now routinely used to produce various sorts of micro- transducers into compact arrays. electromechanical systems, or MEMS. Having a series of transducers laid out in a line— MEMS fabrication techniques have enabled us a one-dimensional array—is the simplest exam- to construct something we call a capacitive micro- ple of this strategy. Such transducer arrays are now machined ultrasonic transducer. This name, we admit, employed routinely for most forms of ultrasound is an ungainly mouthful, and the acronym we use in imaging. Like multielement radio antennas, such our scholarly papers, CMUT, is a bit cryptic to all but arrays can be steered so as to send energy in a narrow, a few specialists. Perhaps this is why some of our col- -yakub directed beam. Steering an array also works in reverse, leagues in industry refer to this new technology by I

allowing it to detect acoustic echoes that come from the more pleasing phrase “silicon ultrasound,” which . khur t one particular direction. While a one-dimensional tells you right away what stuff these new transducers

transducer array can be steered and focused within a are for the most part made of. Continued on page 52 butrus

46 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org 48 NA • iEEE Sp iEEE • NA Laptops, workstations, PlayStations, iPhones—they would all be impossible be all would iPhones—they PlayStations, workstations, Laptops, which hold all his ideas about dynamic dynamic about ideas his all hold which DRAM, and he will probably do three things. He drawwill the circuit fordiagram the one- under a wall crowdedunder a awards. his wall with neers, rather than Dennard and IBM, should ing you how he had the revolutionary idea of random-access memory, meticulously dated transistor DRAM, including every amplifier, ture Cheerios, each of which stored one bit one stored which of each Cheerios, ture Ask Robert Dennard Robert Ask based on the polarity of its magnetic field. field. magnetic its of polarity the on based minia like rings, used—magnetic being books IBM encouraged its inventors to keep, online article that suggests that Intel engi Intel that suggests that article online then technology memory the for capacitor and witnessed by other people, “to make make “to people, other by witnessed and data line, and inverter. and line, data substituting a single transistor and a single a and single transistor a single substituting stores theseHe pristine notebooksinventions.” our in an of armoire proof had we sure Thanks for the E ctrum • • ctrum First, he will show you the patent note patent the you show will he First, Finally, he will comment on a certain certain a on comment will he Finally, Second, Second, he spend will half an hour show mA y 2009 y without without about the invention of Robert Dennard Robert medal medal By

Sally Sally - - - - of honor And he said, ‘We don’t care about inven about care don’t ‘We said, he And worked on the chip, ‘Did you invent DRAM?’ IEEE Medal of Honor. Intel released a thre a released Intel 1970, three years after Dennard entered the Dennard after years three 1970, RAM RAM in these devices might even be taken as pretty inhabits memory Random-access cometo first were theythebecause DRAM, he says. Just about everywhere else, Dennard ing through it: your laptop, cours car, game electrons console, has that everything much tempest. teapot irrelevant hopelessly no is credited as the father of DRAM, and for that pauses.“A lot of people Intel think invented no end: “They asked someone from Intel who tions. We care about products.’ about care We tions. book. The misattribution annoys Dennard to DRAM. of invention the with credited be outsomethingwith labeled dynamicRAM,” one achievementhe is being awarded this year’s digital camera, and cellphone. The amount of a ’s invention of of invention ’s dee The wrestling over who gets credit is is credit gets who over wrestling The - transistor DRAM into his patent note patent his into DRAM transistor Memories DRAM e -transistor DRAM in in DRAM -transistor

” Dennard Dennard ” - - - www.spectrum.ieee.org

DaviD Yellen

a kind of shorthand for their approximate level of performance. That’s because ever-increasing memory capacity is one of the key factors driving the evolution of most electronics. is now a large extended family— including EEPROM (electrically erasable programmable read- only memory) and NAND flash—each category with different drawbacks and benefits. But dynamic random-access memory is an important ancestor. “Random access” means what it says: A microprocessor can withdraw any stored “word” (8 bits of data) from this memory in any order. In Dennard’s one-transistor DRAM, each bit of data is stored separately inside its own capacitor. A single transistor controls both reading and writing. A charged capacitor means “1,” and an uncharged capacitor means “0.” The word dynamic in the name derives from the fact that the act of reading the bit discharges it and it must be rewritten back into memory. A capacitor’s charge eventually wanes, so the memory must be reinfused with fresh charge several times per second to prevent it from losing informa- tion. That fact led one researcher to joke that Dennard had won prestigious awards not for his invention but rather for having the robert dennard temerity to refer to such a thing as “memory.” Amazingly, in an industry defined by its constant advances and compulsory forward movement, the one-transistor DRAM Most Recent AwARDs: Charles Stark Draper Prize, has endured for 40 years. IEEE Medal of Honor DAte of bIRth: 5 September 1932 In 1958, when Dennard walked into IBM’s still-unfinished Thomas J. Watson Research Center for his first day at work, he bIRthplAce: Carthage, Texas didn’t know exactly how a transistor worked. In those days, fAMIly: wife, Jane Bridges, software consultant not too many engineers actually did. Dennard, fresh out of and teacher; two adult daughters from a previ- Pittsburgh’s Carnegie Institute of Technology (now Carnegie ous marriage, each with two children Mellon University), had just earned a Ph.D. in electrical engineer- ing after completing undergraduate and master’s work in EE at pets: Two Scottish terriers, Southern Methodist University, in Dallas. Bonnie and Ferguson But what he recalls most fondly is his first educational expe- rience, one from a bygone era. “At the National Inventor’s Hall of fAvoRIte leIsuRe ActIvIty: Scottish country danc- Fame, a bunch of us guys—all very successful—were having a con- ing, two nights a week, and choral singing versation, and we found out that all of us went to one-room school- MAntRA: “Attitude is everything.” houses,” he says. “That was the common denominator.” Growing up in a 5000-person farm community near the cuRRent tItle: IBM Fellow Louisiana border of Texas, that’s all there was. No Baby Einstein fAvoRIte MovIe: “I don’t watch movies. They’re classes for Dennard, no Mozart symphonies on a phonograph. too loud. The last one I loved was La Ronde, The Depression was just ending; his community hadn’t been elec- which I saw in graduate school 55 years ago.” trified. “We survived just fine,” he says, adding that the secret to his success was that he had a lot of spare time as a child. “I learned everything very slowly and concentrated deeply,” he recalls. ence until he took physics classes at SMU, which he attended In those days, he wasn’t interested in science or engineering on a dual academic/band scholarship as a French horn player. at all. “I had a crystal radio,” he declares, “but I never got that He liked his physics classes, particularly the emerging field of thing to work.” What he loved was science fiction; he devoured semiconductor physics—so much so that he decided to pursue old anthologies that included authors like Edgar Rice Burroughs a doctorate in electrical engineering, which was then an inter- and H.G. Wells. “One story really influenced me,” he recalls. “It esting discipline that in some ways hadn’t quite found itself. was about probability.” The short story, “Inflexible Logic,” by “I had some advanced physics courses, solid-state materials, Russell Maloney, was published in 1940. To test the theory that and so forth,” he says, “but I still didn’t understand exactly patterns would emerge out of randomness, a man assembled six how a transistor operated.” Armed with his Ph.D., he followed monkeys and set them to typing, to see if they would come up some friends to IBM, which was on a research-scientist hir- with anything rational or intelligible. After quite a short time, ing binge. He figured he’d stay for a few years. Fifty-one years the monkeys began to write some very familiar prose. The man later, he’s still there. shared the results with his friend, a professor. He started as a staff engineer in the applied research “And the monkey was coming up with great stuff, and group, studying what were then brand-new metal-oxide- [the professor] was walking around scratching his head and semiconductor field-effect transistor (MOSFET) designs thinking, It couldn’t have happened so soon.” Dennard pauses and circuit applications. Then one day in the fall of 1966, he Yellen

and laughs uproariously. “So he shoots the monkey!” attended an internal IBM Research review conference. One D

Science fiction was as close as he got to an interest in sci- project was an attempt to commercialize magnetic core Davi

50 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org memory, the standard at the time. The magnetic rings were Dennard’s window view is a bit of a cheat. He doesn’t actually strung together on a mesh of wire, forming a grid perhaps have a window. But his door does open directly onto the magnifi- 30 centimeters on a side. “The truth is, it probably wouldn’t cent corridor; his is one of only three offices with that luxury. have worked,” Dennard says. “But it looked good. It was still For all the spartan egalitarianism, Saarinen designed the big, but they had put lots of bits in there.” offices in a bright spectrum of cheery colors. Dennard’s office is Dennard went home that night wondering if he could replace as brash and upbeat as a Piet Mondrian print. Big color blocks the magnetic ring with a small capacitor to store charge. So of built-in filing cabinets cover an entire wall. Dennard spent for the next couple of months he worked on the problem every the majority of his career in a blue office buried in the center day and every night. “The first thing I did was put a transis- of the building; the first-among-equals office he has now is a tor in series with a capacitor. Then you could write the charge happy lime green. into the capacitor and turn it off.” But how to read it? After The IEEE Edison Medal hanging on the wall behind his months of hair pulling, Dennard was seized with his great desk squeezes in next to a row of IBM awards, which in turn eureka moment: a single field-effect transistor and data line rub elbows with a Lemelson-MIT Lifetime Achievement could both read and write the charge stored in the capacitor. Award. On the armoire beneath, haphazard stacks of plaques So in 1967 he detailed his invention in his standard-issue IBM suggest that at some point Dennard gave up the 40-year jigsaw patent notebook, and so was born the one-transistor dynamic puzzle of fitting all the honors onto a single wall. random-access memory. Soon Dennard will need to reorganize again to make room for his 2009 additions: the Charles Stark Draper Prize—an DRAM, like almost all great inventions, has many fathers. annual US $500 000 award conferred by the National Academy Three years before Dennard drew his circuit diagram in of Engineering—and the IEEE Medal of Honor. “They’re not his notebook, fellow IBMers Arnold Farber and Eugene making it any easier,” he says, laughing, as he examines his Schlig had created a memory cell with two transistors and favorite, the heavy bronze National Medal of Technology two resistors. A year later, in 1965, IBM researchers refined awarded to him in 1988 by President Ronald Reagan. that idea into a 16-bit monolithic memory array. Also that “We just didn’t imagine how far it would go,” he says, of the one- year, J.D. Schmidt developed a semiconductor random-access transistor cell, “how much it would totally change computing.” memory, but he used six per memory cell, inflating both the footprint and power consumption. Dennard’s pat- the wall opposite the awards is almost completely filled by a ent for a one-transistor DRAM was awarded in 1968, but IBM chalkboard that hasn’t been erased in months, or maybe years. didn’t turn it into a product. Instead it shipped computers Its runes are of different sizes, with some squeezed into the that used six-transistor SRAM, a technology the company spaces between previous scribblings. There is a small patch considered less risky. of equations with signs for high-k metal dielectrics. In another Then, in 1970, Intel released the first commercially available corner, barely visible under some fresher chalkings, is an equa- DRAM memory chip, the three-transistor 1103, which could tion for measuring capacitance. Dennard preserves them all as store 1024 bits [see “25 Microchips That Shook the World,” in artifacts of the part he likes most about his role at IBM, which this issue]. But that’s not what people mean when they speak is mentoring incoming employees. of DRAM today, Dennard insists. “That’s why they don’t call it “It’s not official mentoring; it’s more like being a professor at ‘one-transistor DRAM,’ ” he says. “It’s just DRAM.” a university,” he explains. “I work with the new people. I work Dennard also conceived the scaling theory of MOSFETs, with them on projects, helping define them, monitor progress, which predicted that the speed of any chip would increase and develop the people. Some of them like it. Others want to in direct proportion to the decrease in size of its transistors. stay well away,” he laughs. This theory is commonly—and erroneously—folded into Ghavam Shahidi, who is also an IBM Fellow, says he bene- Moore’s Law, which actually predicted only the continuing size fited from Dennard’s perspective when he started at the com- decreases, not the associated performance increases. pany as a postdoc 20 years ago. “I knew of him for years before “Bob Dennard was the person who correlated scaling with per- I came to IBM,” he says. “I only knew of his accomplishments, formance, and it’s as important as DRAM,” insists Juri Matisoo, and that was very intimidating at first. I thought, Here’s the who worked on magnetic memory at IBM in the 1960s before man who invented DRAM. This guy is famous.” But Shahidi Dennard wiped out the competition with semiconductor mem- says Dennard was so approachable, down-to-earth, and hum- ory. Matisoo went on to become vice president of technology at ble that the impression did not last. “He was not the way I imag- the Semiconductor Industry Association. “Moore was projecting ined him at all.” the timescale; the IBM people described how to actually do it.” Shahidi, who is credited with the development of silicon-on- insulator semiconductor technology at IBM, says many of his the sprawling t.J. watson research complex in Yorktown epiphanies were born out of long talks with Dennard. “He’s Heights, N.Y., is 61 kilometers north of New York City, but it great to sit down with and just throw out ideas. He has a broad borrows the city’s grid layout, with 40 numbered alleys on perspective that he applies to narrow problems.” each of its three levels. You can’t get lost. The Watson campus Dennard applies his perspective liberally, including to the architecture, finished in 1961, rejected the caste system of cor- recurring “the end is near” refrain that plagues the semicon- porate ambition: No offices have windows. Instead, architect ductor industry. “The first paper warning of the end of scaling Eero Saarinen crafted an enormous communal corridor with a was published by RCA before I even got into this business,” he three-story wall of windows overlooking rolling hills and leafy grouses. “You can always find a reason things can’t be done.” greenery in the summer, bare branches under ice and snow in “That’s the thing about the future,” he exclaims. “It’s totally winter. That bucolic view is available to Nobel Prize winners unexpected. It’s been the same for 50 years—we could never and postdocs in equal measure. see anything more than three years down the road.” o www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 51 In a way, it’s surprising that this approach has taken so Next-Gen Ultrasound long to catch on. After all, condenser microphones are capaci- Continued from page 46 tive sound transducers, and they’ve been common for decades. They change sound into electrical signals using a flexible mem- ltrASoNic trANSducErS have traditionally been brane separated from a solid back plate by a very thin air gap. U fashioned from piezoelectric materials like quartz or lead Both membrane and back plate are conductive, or have conduc- zirconate titanate, which many engineers know as just PZT. These tive electrodes attached to them, so a condenser mic is essen- are crystals or ceramics that expand or contract in response to an tially a parallel-plate capacitor. When sound waves hit the applied voltage. Likewise, piezoelectrics will generate an electric membrane, it vibrates, inducing an oscillatory current from signal in response to being stretched or squeezed, so they can the capacitor when it is biased with a dc voltage. both transmit sound and detect it. This is very old technology, Condenser microphones can capture sound of superb qual- having been invented late in the 19th century, when the brothers ity, which is why they are often used in studio recording. In Pierre and Jacques Curie demonstrated piezoelectric effects in ultrasonics, though, the demands are greater than they are what for decades remained a laboratory curiosity. The first real for audio frequencies. For a capacitive design to be as efficient application, for sonar, came in 1917. as the existing piezoelectric transducers, the electric field in The procedure used to process a piezoelectric substance the gap has to be enormous—hundreds of millions of volts per into a transducer or an array of transducers relies mostly on meter. And when subjected to electric fields of that magnitude, age-old manufacturing methods: mixing materials, bonding air tends to break down, forming a conductive arc. So if you them, mechanically dicing the resulting assembly, adding tried running a normal condenser microphone with a bias volt- wiring—that is, a lot of delicate manual labor. The production age high enough to produce such an electric field, you’d soon of ultrasonic transducer probes, which amounts to a global see sparks fly. market of about $1 billion annually, is therefore limited by the Fortunately, the world works differently at small scales. many headaches involved in maintaining high yield and good As you reduce the size of the gap, the electric field required product uniformity in a manufacturing system that depends for air to break down increases. So with so much on sharp eyes and steady fingers. a sufficiently small gap, you can make a To PRobE The capacitive transducers we’ve been pioneering sidestep capacitive transducer—essentially, a tiny FuRTHER such issues. By using photolithography and other fabrication condenser mic—that supports an immense A detailed summary of the techniques of the semiconductor industry, we can make trans- electric field. Such a transducer can be authors’ mEmS- ducer arrays—large or small—with even the most complex extremely efficient. fabrication geometries, and we can do so very precisely and inexpensively. That a small gap can sustain a large techniques electric field has been known for more is published in “capacitive than a century, but it wasn’t until the micromachined Massachusetts Institute of Technology early 1990s that a few researchers took ultrasonic advantage of this fact and began experi- transducers: Put MIt to Work For You menting with capacitive transducers Fabrication for ultrasonics. They struggled, though, technology,” IEEE Transactions using mostly conventional machining on Ultrasonics, Enroll in MIT classes taken and plastic-film membranes (in a few Ferroelectrics, over one or more semesters, cases with micromachined back plates) and Frequency and were unable to overcome the break- Control, vol. 52, no. 12, december full or part time. down issue in their first crude devices. 2005. Then in 1994 one of us (Khuri-Yakub) and the full range As an ASP Fellow, you will: Matthew I. Haller, who was at that point a of the authors’ graduate student in our research group at research in Stanford, began to apply micromachining acoustics is › investigate the latest advances in your field described at and other MEMS techniques to construct › design a curriculum to meet individual and company goals http://www-kyg. the entire transducer. › learn to harness the power of new technologies stanford.edu. › access the full range of MIT courses and resources At the time, our focus was on equip- › build a lifelong network of colleagues ment for nondestructive testing—looking for cracks in the wings of F-18 fighter jets, to be specific. So we intended these first transducers to be used in air (where they worked surpris- Accepting applications for Fall 2009 ingly well). Having previously done a lot of research for the U.S. Classes begin September 8 Navy on sonar, though, we tried out a pair of the new trans- http://advancedstudy.mit.edu/fall ducers underwater for kicks. What we saw knocked us off our [email protected] chairs. When immersed, the new transducers displayed phe- 617. 253.6128 nomenal bandwidth, much better than piezoelectrics. Actually, they showed this stellar performance for all of 15 minutes or so; then they stopped functioning altogether. After a certain amount of cursing and head scratching, we fig- ured out the reason. In our initial designs, the gap between the membrane and the substrate was left open to the outside envi- ronment. That’s fine for use in air, but when these transducers

52 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org Spire IEEE Spectrum half page Ad 4-6-09.pdf 4/6/2009 1:06:50 PM

traditional new technology technology Obtained with Obtained with piezoelectric- capacitive- transducer transducer probe probe

SCanning-ElECtron miCrograph (ColorEd) Silicon membrane

Evacuated gap Silicon nitride Silicon substrate Silicon dioxide post layer 1 micrometer

nEw VIEw medical ultrasound equipment traditionally uses piezoelectric transducers, usually made from lead zirconate titanate. the upcoming generation of capacitive micromachined transducers offers greater bandwidth, which translates into better depth resolution in the images that can be obtained. the improved quality can be seen in a pair of images of the carotid artery and thyroid gland [top panels], which show anatomical detail better when capacitive transducers are used to obtain the image [right]. Such a transducer consists of many individual cells, each containing a silicon membrane separated from the silicon substrate by a thin gap and a combination of silicon dioxide and silicon nitride insulation [bottom].

C

were immersed, water slowly made its way into the gap, ruinM - ing their ability to operate. But it didn’t take us long to figure out Y how to seal these cavities. And eventually we devised ways to eliminate the air inside altogether. CM Further work also showed why these capacitive transducersMY have greater bandwidth than piezoelectrics. The differenceCY

arises because a piezoelectric transducer is by nature a highlyCMY tuned device, like the pendulum of a clock. At its particular K resonant frequency, a piezoelectric transducer undergoes high- amplitude oscillations, even with very little forcing, but at other frequencies, it barely moves at all—which is to say that it has very limited bandwidth. A capacitive transducer also has a distinct resonant frequency, but only when it’s operating in air. When it’s immersed in water—or coupled to biological tissues, which are much like water in their acoustic properties—the situa- tion becomes very different. Because the vibrating membrane has so little mass, its movements become highly damped by the watery medium it touches. The same thing happens if you place a pendulum under water. It’ll no longer oscillate at its normal resonant frequency, but it can still swing back and forth at the frequency you’re using to drive it. This effect, then, lets a single transducer work well over a broad swath of the to

P ultrasonic spectrum. : da :

VI That’s important because it means that the transducer is d m. m m. d able to emit and detect the many different frequencies that are I lls/G contained in a short ultrasonic pulse. The shorter the pulse you e (2); bottom:t. (2); butrus khur use to probe the patient’s body, of course, the better the depth resolution in the resulting image. And improved resolution is, after all, just what the doctor ordered.

E coNStruct one of our transducers by connecting Wmany small units in parallel. Each contains a thin mem- I -yakub brane, separated from an underlying substrate by a tiny gap. In our latest designs, the membrane is made of silicon, possi-

www.spectrum.ieee.org mAy 2009 • iEEE SpEctrum • NA 53 bly covered with a metal electrode. Silicon ivEN thE many wonderful things is desirable for several reasons. One is Gwe’ve said about them, you might that it has good mechanical properties— think that capacitive micromachined it doesn’t fatigue, for example—and as ultrasonic transducers would already long as it is thin, it will flex sufficiently. be in use in medical imaging equipment. The substrate is silicon as well, doped Many of the companies that make these with a sprinkling of other atoms to make systems have indeed embraced this tech- Visit our website for thousands of standard products it highly conductive. nology, but it hasn’t yet reached vendors’ We’ve developed different recipes for shelves. Most of the remaining techni- www.picoelectronics.com making these devices over the years, but cal issues are minor, though. Some stem TRANSFORMERS & the best scheme we’ve found uses two from the electric fields these transducers Miniaturized INDUCTORS different wafers: a garden-variety sili- must contain. From 1/4" X 1/4" • Surface Mount con wafer for the substrate and one that’s Although there is no chance of arc- Plug-in • Toriodal • Insulated Leads slightly more exotic for the membrane, ing across the evacuated gaps, the enor- something known in the semiconductor mous electric fields can stress the insu- SURFACE Plug-in Mount industry as silicon-on-insulator. The two lating layers to the breakdown point. are bonded together using nothing more And even without that, these large than a modest amount of pressure and fields can inject static electric charge heat. This two-wafer approach permits into those layers, which reduces the DC-DC CONVERTERS Low Profile us to add the membrane after the pockets electric field in the gap, making it nec- Surface Mount • PC Board Mount Single and Dual Output • Up to 10,000V Std. that serve as the gaps are already formed, essary to keep adjusting the dc bias so we can sculpt the membrane and sub- field to compensate. SURFACE Plug-in Mount strate as we wish—they don’t have to be Another challenge with capacitive just flat planes. transducers is that they do not respond Building transducers from silicon as linearly to drive voltages as PZT makes it a snap to connect them with transducers do. Nonlinearity of the the front-end electronics of an ultra- transducer becomes an issue when an DC-DC CONVERTERS High Power sound imager. Although it’s possible to ultrasound system is used to image the Regulated, Up to 400 Watts Up to 100 volts Standard fabricate a transducer directly over the nonlinear response of biological tissues. INdustrial Military associated electronic components on the Fortunately, there are ways to circum- very same silicon wafer, doing so cre- vent this problem, such as purposefully ates a number of troublesome compli- distorting the drive signal to compensate cations. The better tactic, we’ve found, for the nonlinearity of the transducer. is to bond the finished transducer array We—along with a slew of engineers at DC-DC CONVERTERS High Voltage Up to 10,000 VCD Output to a separate wafer containing the elec- Canon, General Electric, LG Electronics, Up to 300 Watts tronic circuitry. National Semiconductor, Siemens, and NEW Dual Outputs New Regulated Output Connecting each transducer ele- elsewhere—are working to solve these ment may be tricky for tightly packed nagging problems and to confront the 2-D arrays, because there isn’t much many other practical realities you have free real estate on the front surface to to deal with in any new product. That’ll route a lot of electrical leads. But here take some time, but it’s clear to us that again the microelectronics industry there are no showstoppers here. AC-DC POWER SUPPLIERS has a good solution: Make the connec- It won’t be long before this new Linear • Switches • Open Frame tions to the electronics by burrowing breed of transducers arrives at hospi- Low Profile • Up to 200 Watts down through the transducer substrate tals all over the world. So expect those and creating vertical conductive chan- first baby pictures you’re shown, among nels, which are known in the trade as other sorts of ultrasound images, soon to through-silicon vias. become even more stunning. o

POWER FACTOR CORRECTED MODULES Universal Input • 47-440 hz Stay Connected to 1000 Watts • 99 Power Factor Spend less time worrying and more time communicating with an IEEE Email Alias.

Call toll free 800-431-1064 for PICO Catalog Fax 914-738-8225 www.ieee.org/alias PICO Electronics,Inc. 143 Sparks Ave, Pelham, NY 10803-1837 www.spectrum.ieee.org E-Mail: [email protected] 401-Qw CS Ad 4.5x2.indd 1 8/10/06 10:30:10 AM

CYAN prints as Pantone 3268C the data

AnnuAl Job Cuts by seCtor electronics telecom computer Dot-com Automotive total u.s. 700 000 unemployment by age feBrUary 2008 To feBrUary 2009 25 600 000 sept. 2002 the 55 and over sept. 2008 Feb. nAsDAQ-100 2009 sinks to 805, from lehman a high of 4705 Brothers collapses, 500 000 in march 2000. 20% triggering a 20 45 to 54 wholesale Feb. 2008 restructuring of the u.s. 400 000 banking system. June 2003 the 35 to 44 15% u.s. Federal reserve 15 makes the first of 17 consecutive increases Aug. 2007 the 300 000 to its discount rate, u.s. Federal reserve 25 to 34 from 2.25 to 6.25. makes the first of 12 consecutive 10% cuts to its discount 45 to 54 10 45 to 54 rate, taking it from 45 to 54 200 000 45 to 54 6.25 to 0.50. 4520 to to 24 54 45 to 54 35 to 44 35 to 44 35 to 44 35 to 44 35 to 44 35 to 44 25 to 34 100 000 5% 25 to 34 5 2516 to to 19 34 25 to 34 25 to 34 25 to 34 20 to 24 20 to 24 20 to 24 perceNTage 20 to 24 20 to 24 INcrease 20 to 24 16 to 19 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009* 16 to 19 16 to 19 16 to 19 16 to 19 20 to 24 16 to 19 16 to 19 25 to 34 0 135 2to 443 4 5 6 High-Tech Unemployment 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 145 2to 543 4 5 6 0 1 2 3 4 5 6 Is Up, But Not Way Up 55 and older 55 and older 55 and older 55 and older nemployment in the united states an employment consulting firm inc hicago. the 55 and older proporTIoNal 55 and older is rising as the current recession cuts 156 000 jobs lost in 2008 were a far cry from INcrease 4555 toand 54 older 45 to 54 deeper than past downturns. some key 2001’s high-water mark of 696 000. compare 45 to 54 u +31%+5.1% 45 to 54 technology sectors are certainly feeling a share that to the automotive industry, where the job 45 to 54 16 to 19 3545 to 4454 of the pain, but through 2008, their situation market is heading over a cliff. 35 to 44 35 to 44 wasn’t all that bad when compared with of course, these figures on job losses don’t +43%+3.9% 35 to 44 20 to 24 35 to 44 unemployment in previous recessions or even in give a full picture of unemployment, because 2535 to 3444 25 to 34 other economic sectors. they don’t take into account the new positions 25 to 34 +3.8%+78% 25 to 34 in the first quarter of this year, however, being filled.s till, challenger spokesperson James 25 to 34 25 to 34 2025 to 2434 high-tech unemployment spiked dramatically pedderson says, the data offer a good picture of 20 to 24 +3.2% +89% 20 to 24 compared to 2008. if extrapolated for the full how things will be going in the next few months. 20 to 24 35 to 44 20 to 24 year, 2009’s totals would be the highest in six pedderson also notes that unemployment 1620 to 1924 16 to 19 years, although still only half that for 2001. doesn’t hit all age groups equally. the overall +2.8% +82% 16 to 19 16 to 19 last year’s job losses for computer, u.s. unemployment rate has almost doubled 45 to 54 16 to 19 electronics, and telecommunications, taken for workers 35 to 44, but it is still far higher 16 to 19 +2.4% +75% together, were up 50 percent compared with among the young. pedderson expects the new 0 20 40 60 80 100 055 and20 older40 60 80 100 the year before—but 2007’s numbers were crop of college graduates to feel the most pain, 0 20 40 60 80 100 0 20 40 60 80 100 unusually low, according to data provided to with both internships and regular jobs hard to 0 20 40 60 80 100 IEEE Spectrum by challenger, gray & christmas, come by. Feb. 2009 —Steven Cherry 0 20 40 60 80 100

*Q1 2009 figures are actual; 2009 totals are projected. SOurcES: challenger, Gray & christmas, New york Federal reserve Bank, http://www.newyorkfed.org/markets/statistics/dlyrates/fedrate.html, yahoo Finance, http://tinyurl.com/cboecl

64 NA • iEEE SpEctrum • mAy 2009 www.spectrum.ieee.org