Innovative Technology for Computer Professionals
January 2004
IEEE Computer Society President’s Message
Design Fragility
Hybrid CPU/FPGA Chips http://www.computer.org
Innovative Technology for Computer Professionals
January 2004,Volume 37,Number 1 PERSPECTIVES 30 A Tale of Three Disciplines ... and a Revolution Jesse H. Poore Some disciplines, like circuit and genetic engineering, seem to evolve from theory to practice relatively responsibly. Software engineering, on the other hand, while not yet at the guillotine, has suffered a decided lack of direction. It may be time to storm the gates. 37 The End of Science Revisited John Horgan Because science has advanced so rapidly over the past century or so, we assume that it can and will continue to do so, possibly forever. But science itself tells us that there are limits to our knowledge. COVER FEATURES 47 2003 Technology Roadmap for Semiconductors Don Edenfeld, Andrew B. Kahng, Mike Rodgers, and Yervant Zorian This update to the 2001 ITRS Roadmap shows the industry shifting its focus toward systems on chip, wireless computing, and mobile applications. 59 NASA’S Mission Reliable Patrick Regan and Scott Hamilton Like industrial development organizations, the US space agency struggles with the challenge of creating reliable software. NASA’s deep space community is attacking its software crisis via two complementary approaches—one stressing the power of engineering discipline, the other the potential of automated code generation and Cover design and artwork by Dirk Hagner verification. 69 IT Employment Prospects in 2004: A Mixed Bag ABOUT THIS ISSUE Fred Niederman n this January Outlook issue, Signs of general economic recovery and renewed IT sector growth we continue our tradition of are dampened for IT workers globally by improved productivity and I publishing articles that locally by a global labor market. address advances with near-term possibilities. 78 DakNet: Rethinking Connectivity in Developing Topics covered include the ITRS assessment of Nations the challenges awaiting the semiconductor Alex (Sandy) Pentland, Richard Fletcher, and Amir Hasson industry; NASA’s software reliability efforts at DakNet provides extraordinarily low-cost digital communication, JPL and the Ames Research Center and its letting remote villages leapfrog past the expense of traditional systems and software architecture for next- connectivity solutions and begin development of a full-coverage generation unmanned spacecraft; an innovative broadband wireless infrastructure. initiative for providing inexpensive wireless connectivity for rural populations in developing 85 Combining Optical Holograms with Interactive nations; an assessment of global IT employment Computer Graphics prospects in 2004; and a novel system for Oliver Bimber combining optical holograms with interactive Merging optical holograms with 3D graphical elements can provide computer graphics. This issue also includes two an acceptable tradeoff between quality and interactivity: The essays on the future of software engineering and holographic data provides high-quality but static content, while the end of science. additional graphical information can be generated, inserted, modified, and animated at interactive rates. IEEE Computer Society: http://computer.org Computer: http://computer.org/computer [email protected] IEEE Computer Society Publications Office: +1 714 821 8380 OPINION 13 At Random Design Fragility Bob Colwell NEWS 18 Industry Trends Vendors Go to Extreme Lengths for New Chips Steven J. Vaughan-Nichols 22 Technology News Is Peer-to-Peer Secure Enough for Corporate Use? George Lawton 26 News Briefs E-Mail Security Gets Personal Ceiling for Controversial US H-1B Visas Drops Interactive Panoramic Images MEMBERSHIP NEWS 8 EIC’s Message 10 President’s Message 92 Call and Calendar 95 Computer Society Connection COLUMNS 116 Web Technologies Databases Deepen theWeb Thanaa M. Ghanem and Walid G. Aref 118 Embedded Computing Programming Models for Hybrid CPU/FPGA Chips David Andrews, Douglas Niehaus, and Peter Ashenden 121 IT Systems Perspectives Distributed Learning Environments Maryam Alavi 124 Standards Best Business Practices for Standards Groups NEXT Michael LoBue MONTH: 128 The Profession Ad Hoc From Organization Man to Free Agent Networking Mohan Babu DEPARTMENTS 4 Article Summaries 6 Letters 17 32 & 16 Years Ago 44 IEEE Computer Society Membership Application 100 Career Opportunities 113 Advertiser/Product Index 114 Products 115 Bookshelf
COPYRIGHT © 2004 BY THE INSTITUTE OF ELECTRICAL AND ELECTRONICS ENGINEERS INC. Membership Magazine ALL RIGHTS RESERVED. ABSTRACTING IS PERMITTED WITH CREDIT TO THE SOURCE. LIBRARIES of the ARE PERMITTED TO PHOTOCOPY BEYOND THE LIMITS OF US COPYRIGHT LAW FOR PRIVATE USE OF PATRONS: (1) THOSE POST-1977 ARTICLES THAT CARRY A CODE AT THE BOTTOM OF THE FIRST PAGE, PROVIDED THE PER-COPY FEE INDICATED IN THE CODE IS PAID THROUGH THE COPYRIGHT CLEARANCE CENTER, 222 ROSEWOOD DR., DANVERS, MA 01923; (2) PRE-1978 ARTICLES WITHOUT FEE. FOR OTHER COPYING, REPRINT, OR REPUBLICATION PERMISSION, WRITE TO COPY- RIGHTS AND PERMISSIONS DEPARTMENT, IEEE PUBLICATIONS ADMINISTRATION, 445 HOES LANE, P.O. BOX 1331, PISCATAWAY, NJ 08855-1331. Innovative Technology for Computer Professionals
Editor in Chief Computing Practices Special Issues 2004 IEEE Computer Society Doris L. Carver Rohit Kapur Bill Schilit President Louisiana State University [email protected] [email protected] Carl K. Chang [email protected] [email protected] Perspectives Web Editor Associate Editors Bob Colwell James H. Aylor in Chief [email protected] [email protected] Bill Schilit Intel Research Features Kathleen Swigger Kathleen Swigger University of North Texas [email protected]
Area Editors Column Editors Databases/Software At Random The Profession Edward A. Parrish Michael Blaha Bob Colwell Neville Holmes Worcester Polytechnic Institute OMT Associates Inc. Bookshelf University of Tasmania Ron Vetter University of North Carolina at Wilmington Multimedia Michael J. Lutz Security Alf Weaver Savitha Srinivasan Rochester Institute of Bill Arbaugh University of Virginia IBM Almaden Research Center Technology University of Maryland Networking and Communications Standards CS Publications Board Multimedia Upkar Varshney Gary Robinson Michael R. Williams (chair), Jean Bacon, Jonathan Liu Georgia State University EMC Corp. Laxmi Bhuyan, Mark Christensen, University of Florida Embedded Computing Web Technologies Thomas Keefe, Deepandra Moitra, Steven L. Tanimoto, Anand Tripathi Software Wayne Wolf Sumi Helal H. Dieter Rombach Princeton University University of Florida CS Magazine Operations AG Software Engineering Entertainment Computing Advisory Panel Committee Dan Cooke Michael R. Macedonia James H. Aylor Bill Schilit (chair), Jean Bacon, Pradip Bose, Texas Tech University Georgia Tech Research Institute University of Virginia Doris L. Carver, George Cybenko, John C. Dill, Frank E. Ferrante, Robert E. Filman, IT Systems Perspective Thomas Cain Richard G. Mathieu University of Pittsburgh Forouzan Golshani, David Alan Grier, St. Louis University Rajesh Gupta, Warren Harrison, Ralph Cavin Mahadev Satyanarayanan, Semiconductor Research Corp. Invisible Computing Nigel Shadbolt, Francis Sullivan Bill Schilit Ron Hoelzeman Intel University of Pittsburgh
Editorial Staff Administrative Staff Scott Hamilton Mary-Louise G. Piner Executive Director Sales Representative Senior Acquisitions Editor Staff Lead Editor David W. Hennage Sandy Brown [email protected] Bob Ward Publisher Advertising Coordinator Judith Prow Membership News Editor Angela Burgess Marian Anderson Managing Editor Bryan Sallis [email protected] Manuscript Assistant Assistant Publisher James Sanders Design Dick Price Senior Editor Larry Bauer Membership & Circulation Linda World Dirk Hagner Marketing Manager Senior Editor Georgann Carter Lee Garber Production Senior News Editor Larry Bauer Chris Nelson Associate Editor
Circulation: Computer (ISSN 0018-9162) is published monthly by the IEEE Computer Society. IEEE Headquarters, Three Park Avenue, 17th Floor, New York, NY 10016- 5997; IEEE Computer Society Publications Office, 10662 Los Vaqueros Circle, PO Box 3014, Los Alamitos, CA 90720-1314; voice +1 714 821 8380; fax +1 714 821 4010; IEEE Computer Society Headquarters,1730 Massachusetts Ave. NW, Washington, DC 20036-1903. IEEE Computer Society membership includes $14 for subscription of Computer magazine ($14 for students). Nonmember subscription rate available upon request. Single-copy prices: members $10.00; nonmembers $20.00. Postmaster: Send undelivered copies and address changes to Computer, IEEE Service Center, 445 Hoes Lane, Piscataway, NJ 08855. Periodicals Postage Paid at New York, New York, and at additional mailing offices. Canadian GST #125634188. Canada Post Publications Mail (Canadian Distribution) Agreement Number 0487910. Printed in USA. Editorial: Unless otherwise stated, bylined articles, as well as product and service descriptions, reflect the author’s or firm’s opinion. Inclusion in Computer does not necessar- ily constitute endorsement by the IEEE or the Computer Society. All submissions are subject to editing for style, clarity, and space.
2 Computer ARTICLE SUMMARIES
A Tale of Three Disciplines ... systems for wireless communication and jobs. For US workers, this change raises and a Revolution embedded applications. Battery-pow- two different but complementary ques- pp. 30-36 ered mobile devices are replacing wall- tions. First, what are the prospects for the Jesse H. Poore plugged servers. SoC and system- global IT workforce in the near and in-package designs are supplanting in- longer term? Second, how will IT jobs be espite years of hard work and ded- house, single-source chip designs. distributed among competing labor mar- icated champions, software engi- Software today can account for 80 per- kets around the world? D neering’s educational and indus- cent of an embedded system’s develop- trial constituents have not only failed to ment cost, test cost has increased expo- converge, the gap between them contin- nentially relative to manufacturing cost, DakNet: Rethinking Connectivity ues to grow. and verification engineers outnumber in Developing Nations Some disciplines, like circuit and design engineers on microprocessor pro- pp. 78-83 genetic engineering, seem to evolve from ject teams. Overcoming the many exist- Alex (Sandy) Pentland, Richard theory to practice relatively responsibly. ing design technology gaps will thus Fletcher, and Amir Hasson Software engineering, on the other hand, require a concerted effort by the entire while not yet at the guillotine, has suf- industry. hat is the basis for a progressive, fered a decided lack of direction. It may market-driven migration from be time to storm the gates. It is never too W e-governance to universal late for a revolution, and circuit and NASA’s Reliable Software Mission broadband connectivity that local users genetic engineering provide two worthy pp. 59-68 will pay for? role models. Patrick Regan and Scott Hamilton DakNet, an ad hoc network that uses wireless technology to provide asyn- oth predictable and unpredictable chronous digital connectivity, is evidence The End of Science Revisited hazards await the spacecraft, that the marriage of wireless and asyn- pp. 37-46 B robots, and scientific instruments chronous service may indeed be the John Horgan that humans dispatch to explore our beginning of a road to universal broad- solar system. The toughest hazard may band connectivity. DakNet has been suc- cientists need a certain degree of be the known presence of unknown bugs cessfully deployed in remote parts of faith to bolster their confidence in in even rigorously tested software. both India and Cambodia at a cost two S the arduous quest for truth; lack- By exploring new technologies and orders of magnitude less than that of tra- ing such faith, science would not have approaches to develop provably reliable ditional landline solutions. come so far so fast. But when researchers software within tough constraints, NASA reflexively deny any evidence and argu- has a chance to advance the state of the ments that challenge their faith, they vio- art, contributing to computer science as Combining Optical Holograms late the scientific spirit. well as software engineering. In addition, with Interactive Computer Although we have grown up in a any successful spin-off that improves reli- Graphics period of explosive scientific and techno- ability while cutting development time pp. 85-91 logical progress, reflected by such mea- and costs could, in principle, generate sav- Oliver Bimber sures as Moore’s law, science—especially ings for US industry equal to the nation’s pure science—might be entering an era of budget for space exploration. olograms can reconstruct complete diminishing returns. Science itself tells us optical wavefronts, capturing images that there are limits to our knowledge. H that have a three-dimensional ap- IT Employment Prospects in 2004: pearance and can be observed from dif- A Mixed Bag ferent perspectives. Museum exhibits 2003 Technology Roadmap for pp. 69-77 often use optical hologram technology Semiconductors Fred Niederman because it permits presentation of 3D pp. 47-56 objects with almost no loss in visual Don Edenfeld, Andrew B. Kahng, ince 1999—when US business quality. Optical holograms are static, Mike Rodgers, and Yervant Zorian magazine cover stories described IT however, and lack interactivity. Combin- S positions going unfilled and exten- ing 3D computer graphical elements with oday, microprocessor advances sive congressional lobbying to increase stereoscopic presentation techniques pro- abound: general-purpose digital quotas for overseas workers to fill vides an alternative that allows interac- T microprocessors for personal com- them—the US IT job market has changed tivity. puters have been joined by mixed-signal drastically, losing more than a million
4 Computer On the Forefront of Computer Science... The Latest From Wiley-IEEE Computer Society Press
It Sounded Good NEW When We Started A Project Manager's Guide to Working with People on Projects Dwayne Phillips, Roy O'Bryan The authors, both respected project managers with sixty years of experience between them, describe their own mistakes as well as the many valuable lessons they drew from them. Instead of trying to formulate these in abstract theory, Phillips and O’Bryan tell the stories surrounding a particular project, providing Top Selling Titles from a more memorable, real-world, and practical set of examples. Wiley-IEEE Computer Society Press A Wiley-IEEE Computer Society Press Publication ISBN: 0-471-48586-1 Price: $55.00 Software Engineering Pages: 344 Date: Nov. 2003 Merlin Dorfman, Richard H. Thayer (Foreword by Barry W. Boehm) ISBN: 0-8186-7609-4 Price: $63.50 Pages: 552 (Paper) Date: 1997 Multiwavelength Optical LANs Iterative Computer Algorithms with Applications in Engineering Solving Combinatorial Optimization Problems G. I. Papadimitriou , P. A. Tsimoulas, Sadiq M. Sait, Habib Youssef M. S. Obaidat, A. S. Pomportsis ISBN: 0-7695-0100-1 Price: $69.95 Pages: 410 (Paper) Date: 2000 Enables readers to understand the concepts of optical LANs and Digital Systems Design with VHDL and Synthesis learn the details of architecture issues and control protocols. An Integrated Approach The book focuses on major architectural, topological and protocol K.C. Chang issues and presents the important class of adaptive protocols. It ISBN: 0-7695-0023-4 Price: $59.95 Pages: 516 Date:1999 provides specific and extensive coverage of optical LANs ISBN: 0-470-85108-2 Price: $79.95 Pages: 334 Date: Dec. 2003 UMTS Security Valtteri Niemi, Kaisa Nyberg GSM, GPRS and EDGE Performance Focuses on the standardized security features of UMTS and Evolution Towards 3G/UMTS 2nd Edition brings together material previously only available in specifications, design documents and presentations in one concise form. In Timo Halonen, Javier Romero, Juan Melero addition, this unique volume also covers non-standard Fully revised and updated throughout, the book continues to implementation specific features that allow differentiation between provide a complete overview of the entire GSM system. It operators and manufacturers. presents comprehensive descriptions of GSM’s main evolutionary ISBN: 0-470-84794-8 Price: $79.95 milestones – GPRS, AMR and EDGE and explains how such Pages: 286 Date: Jan. 2004 developments have positioned GERAN (GSM/EDGE Radio Access Network) as a full 3G radio standard. 3rd Edition ISBN: 0-470-86694-2 Price: $110.00 Smart Card Handbook Pages: 654 Date: Dec. 2003 W. Rankl, W. Effing Fully revised, this handbook describes the advantages and Next Generation SONET/SDH disadvantages of smart cards when compared with other Voice and Data systems, such as optical cards and magnetic stripe cards and explains the basic technologies to the reader. This book also Stamatios V. Kartalopoulos considers the actual status of appropriate European and Dr. Kartalopoulos once again brings his accessible presentation international standards. of optical communication technologies to SONET/SDH. This ISBN: 0-470-85668-8 Price: $160.00 book introduces communications professionals and students to Pages: 1,120 Date: Feb 13, 2004 the new generation of SONET/SDH and to SONET over DWDM. The material is presented in a highly accessible manner with only essential mathematics and supported with illustrations that Ordering Information enhance understanding of complex topics and procedures. CALL: MAIL: A Wiley-IEEE Press Publication North America: 1-877-762-2974 John Wiley & Sons, Inc. ISBN: 0-471-61530-7 Price: $89.95 Rest of the World +44(0) 1243 779 777 Customer Care-Wiley Pages: 216 Date: Jan. 2004 FAX: 10475 Crosspoint Blvd. U.S. Customers: 1-800-597-3299 Indianapolis, IN 46256 Register for our E-mail Service and Outside U.S. :+44(0) 1243 843 296 E-MAIL: Receive 20% Off your next on-line order. Visit U.S. Customers: [email protected] www.wiley.com/enewsletters to register today! www.wiley.com/ieeecs Outside U.S.: [email protected] LETTERS
THE PROBLEM WITH AI Holmes refers to, there is no conflict between them. They are just different I read Neville Holmes’s column about ways to study memory. One is more artificial intelligence with great interest structure based, while the other ex- (“Artificial Intelligence: Arrogance or plains various memory features. Ignorance?” Nov. 2003, pp. 120, 118- Human intelligence, call it a problem- 119). solving capacity of any kind, uses both The family patriarch, a rocket scien- kinds of knowledge. Basically, human tist by profession, once said, “Before intelligence does not really know which one can have ‘artificial intelligence’, knowledge domain it is using. Gard- one needs ‘natural intelligence.’” As even if only to keep researchers on the ner’s multidimensional view of intelli- Mr. Holmes opined, the implication is same page. gence comes into play here. that the people attempting to imple- Gerad Welch What we can put into a computer ment AI would first have to understand Rochester, Minn. program directly, as software or as the fundamentals of human intelli- [email protected] data, is basically of a declarative nature. gence. It was also a jibe at those who A piece of software—whether AI or employ the noble moniker to describe Neville Holmes responds: conventional—can collect procedural trivial mechanisms. It is perhaps relevant to note that not so knowledge, but this occurs indirectly. The problem with today’s so-called AI long ago, a column called Open Chan- The software must be prepared for that is largely that, as Mr. Holmes describes, nel used to appear on the back page of purpose and must be trained to per- the processes cannot yet adapt algo- Computer. Heading it was a quotation form a specific task, such as face or rithms. While Cellary’s assertion regard- ascribed to Charles McCabe of the San speech recognition. ing the ability to use knowledge to make Francisco Chronicle: “Any clod can This procedural knowledge is also a decisions seems to be on the right track, have the facts, but having an opinion bit restricted by what can be captured I doubt that normal people would con- is an art.” as digital data—unlike human intelli- sider someone who spews facts, yet is gence, which uses all knowledge facil- unable to assimilate new data, as being DEFINING MEMORY ities available. This restriction limits intelligent. a program’s intelligence, as does not The key to having an intelligent In his discussion of AI, Neville Holmes knowing what data we use. process is that the entity has enough does not sufficiently take into account Jan Giezen sensory inputs and meaningful out- the results from current brain research. Delft, Netherlands puts—a feedback loop as per Norbert This research, which uses positron [email protected] Wiener—to validate internally formed emission tomography and magnetic theory on reality. Without the ability resonance imaging, reveals two distinct Neville Holmes responds: to train in a substantially closed- types of memory: declarative (memory) The points that Jan Giezen makes all loop fashion, new things cannot be and nondeclarative (procedural). boil down to a fundamental issue that learned. Essentially, declarative memory is I have raised before: Computing peo- The debasing of the AI concept by what we can express in words or bring ple have much too simple ideas about ascribing the term to applications to mind as a mental image; it is explicit the human brain (“Would a Digital unworthy of the name is why AI pio- or conscious memory. In contrast, Brain Have a Mind?” Computer, May neer Marvin Minsky declared, “AI has nondeclarative memory is the collec- 2002, pp. 112, 110-111). been brain-dead since the 1970s.” The tion of skills we acquire with practice To say that human memory is of software industry has relegated “AI or repetition—our habituation or only two kinds is a gross simplification. technology” to interactive help and conditioning. For example, artists, I haven’t read the book that Giezen video game opponents. Of course, sci- musicians, and athletes are masters of cites, but other books that I have read, entists often use terms that don’t have specific procedural knowledge. As such as Memory: Phenomena and the same meaning to the general public. Larry R. Squire and Eric R. Kandel Principles by Norman E. Spear and Perhaps AI has become one of those explain in Memory: From Mind to David C. Riccio (Allyn and Bacon, words. Molecules (W.H. Freeman, 2000), 1994), emphasize the complexity of I like the term “algoristics” that Mr. much of what Holmes refers to is human memory, and indeed that of Holmes proposes to distinguish static procedural memory. other animals. Popular writings dis- expert systems from the type of intelli- Although these two divisions differ tinguish many other kinds of memory, gent systems we have yet to implement, from Howard Gardner’s independent episodic memory being one that and I hope that others will adopt it— dimensions of intelligence that Neville springs to mind.
6 Computer How to Reach
To equate human memory with Computer intelligence is wrong. Dumb people can have good memories, and smart people can have bad memories. This is Writers where Gardner’s work comes in. There We welcome submissions. For detailed information, visit www. are many quite different and relatively computer.org/computer/author.htm. independent intelligences. It seems to me that an intelligence in Gardner’s sense is a talent for exploiting percep- News Ideas tion and memory to produce high- quality behavior in a particular area. Contact Lee Garber at lgarber@ computer.org with ideas for news features or news briefs. DIVISIBLE LOAD-SCHEDULING DISCOVERY Products and Books Since the publication of my article, Send product announcements to [email protected]. Contact “Ten Reasons to Use Divisible Load Theory” (Computer, May 2003, pp. [email protected] with book announcements. 63-68), I have become aware of an arti- cle by R. Agrawal and H. V. Jagadish published in 1988 (“Partitioning Tech- Letters to the Editor niques for Large-Grained Parallelism,” Please provide an e-mail address with your letter. Send letters to IEEE Transactions on Computers, [email protected]. Dec. 1988, pp. 1627-1634) that, inde- pendently of the earliest work men- tioned in my article (also published On the Web in 1988), discusses divisible load mod- eling. Explore www.computer.org/computer/ for free articles and general Although apparently unknown to information about Computer magazine. many divisible load-scheduling re- searchers until recently, this article has some noteworthy features and firsts. It Magazine Change of Address models the divisible load-scheduling Send change-of-address requests for magazine subscriptions to problem using Gantt-like charts and includes solution reporting time in [email protected]. Make sure to specify Computer. much the same way that others have done since then. This appears to be the first paper to discuss a linear pro- Missing or Damaged Copies gramming solution (unlike the alge- If you are missing an issue or received a damaged copy, contact braic solution discussed in our first [email protected]. paper), a proof on the optimal order for solution reporting, and an experi- mental evaluation of divisible load Reprint Permission scheduling. All in all, it is quite a for- ward-looking paper. To obtain permission to reprint an article, contact William Hagen, Thomas Robertazzi IEEE Copyrights and Trademarks Manager, at [email protected]. Stony Brook, N.Y. To buy a reprint, send a query to [email protected]. [email protected]
We welcome your letters. Send them to [email protected]. Letters are subject to editing for style, clarity, and length.
January 2004 7 EIC’S MESSAGE
Reviewing 2003 Looking back at 2003, our topic cov- Computer erage was diverse. A feature article in our January 2003 Outlook issue pro- vided coverage of NASA’s research in advanced computing technologies to in 2004 help usher in a new area of autonomous space exploration. This issue also in- Doris L. Carver, Louisiana State University cluded an article describing NSF’s TeraGrid computing infrastructure to support collaborative computational science and access to distributed re- sources. Another feature article review- elcome to the January ed advances in integrated circuits and 2004 issue of Computer. MEMS that are enabling neurobiolo- My vision for Computer gists to link computer circuitry to neural W in 2004 is that it will cells in live animals and study the basis continue to serve as a of animal behavior and intelligence. leading information resource in the Other issues published during the lifelong learning process of our read- year highlighted topics such as com- ers. We remain committed to inform- mercial workloads, pervasive comput- ing the computing community of the ing, agile software, piracy and privacy, latest advances in research, technol- Doris L. Carver, nanocomputing, handheld computing, ogy, and applications, as well as pro- Editor in Chief, Computer Web services, safety-critical systems, viding timely information about pro- and power-aware computing. fessional issues, career trends, news, practices, and perspectives on com- ing. We use the results of that session as Looking ahead to 2004 puting-related topics. the framework for developing our edi- In this January Outlook issue we torial calendar for the following year. continue our tradition of publishing COMPUTER’S COVERAGE Potential topics are assessed from articles that address advances with Increasing computational capabili- numerous perspectives, including time- near-term possibilities. ties and decreasing computation costs liness, breadth of coverage, and rele- Topics covered in this issue include combine to escalate the demand for vance to our readers. the International Technology Roadmap complex applications that have far- I would like to clarify the role that for Semiconductors’ assessment of the reaching impacts on society, including the editorial calendar plays in our edi- challenges awaiting the semiconductor new frontiers in medical diagnosis and torial process. In particular, I want to industry, especially as mixed-signal and treatment, weather forecasting, space inform potential authors that we wel- application-driven technologies come exploration, e-commerce, and educa- come submissions on any computing- to the fore. Another feature article tion. Advances in a diverse set of com- related topic. reviews NASA’s software reliability puting areas, such as ubiquitous Our editorial calendar does not efforts at JPL and the Ames Research computing, data integration, self-adap- imply that the topics listed there are the Center and its ambitious systems and tive systems, and multimodal sensors, only areas in which we wish to publish. software architecture for next-genera- are fundamental to realizing these new Special issues focus on a specific topic tion unmanned spacecraft. Other arti- frontiers. As the computing field con- while others are theme issues. Special cles describe an innovative initiative for tinues to thrive and change to meet issues are guest edited by experts in the providing inexpensive wireless connec- new challenges, Computer continues specific area, and theme issues include tivity for rural populations in develop- to evolve to provide coverage of these multiple articles on the theme topic but ing nations, provide an assessment of changes and challenges. are not the result of a specific call. global IT employment prospects in One of these challenges is providing In both special and theme issues, we 2004, and discuss a novel system for coverage of all aspects of computing also include articles on other topics. So combining optical holograms with that our readers find meaningful. We I invite and encourage potential interactive computer graphics. Two initiate the process for determining our authors to submit your manuscripts provoca- tive essays consider the future primary topics with a lengthy session on all computing-related topics to of soft- ware engineering and ponder at our summer editorial board meet- Computer. the end of science.
8 Computer Our 2004 planned coverage in- tributions as a member of the Advisory electronic process for manuscript cludes, but is not limited to, comput- Panel, and to Bill Mangione-Smith for submission and processing. ers and the aging; adaptive software his service as area editor for high- • Indicate your interest in serving as and hardware; Internet security; next- performance computing. Gary, Bill, a reviewer by sending an e-mail generation search; sensor networks; and Jerzy have provided valuable time, message that contains your vitae software architectures; and Internet insights, vision, and guidance to the to [email protected]. data centers. magazine. • Propose a special issue by con- Our column and department offer- tacting Bill Schilit, Special Issues ings will continue to include At We welcome your input Editor, at [email protected]. Random, The Profession, Communi- and diligently strive to use • Provide feedback, including sug- cations, Embedded Computing, Enter- it as part of our ongoing gestions for topics, by sending an tainment Computing, IT Systems e-mail message to dcarver@lsu. assessment process. Perspectives, Invisible Computing, edu. Security, Standards, Web Technologies, • Visit our Web site at www. News, Products, and Bookshelf. We computer.org/computer, where also introduce a new column, 32 & 16, We welcomed two new members to you will find a dynamic, value- that looks back on computing tech- the Board during 2003. Daniel E. added extension of the magazine. nology and the profession through the Cooke joined the Board as an area edi- lens of Computer. tor for software. Dan, a professor and I look forward to your participation If you have suggestions for other chairman of the Computer Science in 2004! topics, please e-mail them to me at Department of Texas Tech University, [email protected]. previously served as the Program APPRECIATION Manager for NASA’s National I wish to thank the members of the COMPUTER AS A Strategic Initiative for Intelligent editorial board and editorial staff for MEMBERSHIP MAGAZINE Systems. Savitha Srinivasan assumed their continued commitment and ded- Computer is the primary communi- the role of area editor for multimedia. ication to achieving Computer’s goals. cation path between the IEEE Com- Savitha is the manager of content pro- They play a crucial role in making puter Society and its members. Through tection at the IBM Almaden Research Computer happen. The volunteer and Computer, we regularly inform you Center, where she defines new research staff partnership that drives Computer about initiatives and events that are areas in content protection and is creates a dynamic, fulfilling environ- happening at the Society, such as actively involved with content protec- ment in which to serve as EIC. the Distance Learning Campus (www. tion standards activities. To our authors and reviewers, please computer.org/distancelearning/); the In 2004, Alfred Weaver is moving accept my deepest gratitude for your Certified Software Development Pro- from his role as area editor of net- contributions. Computer’s value and fessional program (www.computer.org/ working and e-commerce to serve on success lie with authors who submit certification/); the Computer Society the Advisory Panel. I thank Alf for his their work to us and to reviewers who International Design Competition dedication as an area editor and look help ensure that our standards are (www.computper.org/csidc); the distin- forward to working with him as a maintained. The result of our excellent guished recipients of the Computer member of the panel. My thanks to authors and reviewers is that our Society awards; new IEEE Fellows; and Dan, Savitha, and Alf for their will- acceptance rate for articles is approx- volunteer opportunities such as editor- ingness to contribute their expertise to imately 25 percent. in-chief positions for Society publica- Computer. Finally, thank you to our readers. tions. We plan to expand coverage of We welcome your input and diligently Society initiatives of interest to our AN INVITATION strive to use it as part of our ongoing readers in 2004. I invite you to contribute to assessment process. Computer through any of the follow- I hope you enjoy this January EDITORIAL BOARD ing methods: Outlook issue. CHANGES FOR 2004 I would like to express my gratitude • Submit your manuscript for con- to the editorial board members whose sideration for publication at cs- Doris L. Carver is the associate vice term of service ended in 2003. My sin- ieee-manuscriptcentral.com. This chancellor of Research and Graduate cere appreciation to Gary Robinson for site, which contains complete Studies and a professor of computer his service as editor of the Standards author information and submis- science at Louisiana State University. column, to Jerzy Rozenblit for his con- sion details, provides a totally Contact her at [email protected].
January 2004 9 PRESIDENT’S MESSAGE
peer-reviewed magazine on security and privacy (www.computer.org/ Evolving the World security). We also are the first IEEE society to provide our members with free access to an online library of ref- erence books (www.computer.org/ with the World’s bookshelf). While the Society as a whole has accomplished its mission in the past, our latest planning effort convinced us Computer Society that it is time to move forward with an expanded vision and new strategic Carl K. Chang, IEEE Computer Society 2004 President activities. DON’T LEAVE HOME WITHOUT IT: VISA During my presidential election n Computer’s first issue for 2004, campaign, I communicated with you I extend a warm welcome to the about my VISA concept: Vision, nearly 100,000 new and continu- Interoperability, Strategy, and Action. I ing IEEE Computer Society mem- I am happy to report that during my bers. We reaffirm our commitment year as president-elect, I began formu- to offering you those products and ser- lating VISA with strong support from vices that will most benefit you in your volunteer leaders and the your careers as computing and IT Society’s executive staff. As a result of professionals. this collective effort, we have articu- lated a new vision for the Society “to 2003 ACCOMPLISHMENTS Carl K. Chang be the leading provider of technical Under the leadership of Past President IEEE Computer Society President information, community services, and Stephen L. Diamond, our Society made personalized services for the world’s substantial progress in many areas last computing professionals.” year. Our 2003 accomplishments were We decided to adopt interoperabil- outlined in Computer’s December issue. statement “to be the leading provider ity—defined as organizing and collab- On behalf of the Society, I wish to thank of technical information and services orating to deliver an integrated set of Steve for his dedicated efforts to develop for the world’s computing profession- service packages across organizational new initiatives that focused on improv- als.” To accomplish our mission, we units—as the overarching theme for ing the Society’s operational effective- have led the IEEE in product and mem- our planning effort. We then investi- ness. He also guided the advancement ber service innovations. gated ways of building a sound infra- of our ongoing electronic projects, The Society was the first within the structure to foster interoperability including distance learning and the IEEE to initiate an online digital library among all segments of the Society. Total Information Provider project. (www.computer.org/publications/dlib); Much of our effort in defining new Another 2003 initiative was the offer distance learning courses as a strategies centered on the concept of development of the Society’s next strate- basic member benefit (www.computer. a community-driven, service-centric gic plan, SP-5. As president-elect, my org/distancelearning); and launch an infrastructure, with interoperability as responsibility was to lead the effort to e-version-only magazine, IEEE Distrib- the “enabling technology.” The Society rethink our future directions, and I uted Systems Online (http://dsonline. has built or experimented with several engaged all Society Executive Commit- computer.org). online technical communities, most tee members in a forum to create SP-5. We were the first among peer profes- notably DS Online and the IEEE SE This plan builds on a strong foundation sional associations to define the body Online, a new software engineering established by my predecessors. of knowledge of software engineering community to be launched early this (www.computer.org/swebok), to offer year. MOVING PAST THE GLORIOUS PAST a certification program for software Although not all communities are In recent years, the Society’s strategic professionals (www.computer.org/ created equal, they do have some planning has been guided by its vision certification), and to launch a technical common features. A community typi-
10 Computer cally represents a group of individuals dent between the terms of a highly Jim’s goal is to provide a consistent set with common interests who come experienced industrial executive and a of Society offerings serving software together in a common space—in this seasoned university scholar. engineers in their professional devel- case a Web-based environment—to First Vice President Lowell G. opment. In addition, Jim wants to cre- collaborate and to exchange informa- Johnson, an independent consultant, ate models of membership roles to tion. will chair the newly established better structure products to serve our The Society must devise better mech- Electronic Products and Services diverse membership. anisms to continue supporting techni- Board. This board will create new Christina M. Schober, a product cal activities that appear to become dimensions of products and services team leader/staff engineer at Honeywell more fragmented day by day. The com- that capitalize on emerging and mature Aerospace Electronic Systems, will be munity concept was identified as an electronic technologies. Opportunities vice president for conferences and tuto- effective and efficient method to achieve for us to innovate in this area are defi- rials. Chris will continue her efforts to this end. nitely abundant. restructure the conference business By the same token, since our model to create more incentives for resources are limited, the Society can conference organizers. no longer operate in a production- Our vision is for the Murali Varanasi, who is starting an centric mode, such as continuing to Society “to be the leading electrical engineering department at the launch periodicals. We must transform provider of technical University of North Texas and is a our organization into a service integra- information, community long-time volunteer for accreditation tor with less capital investment and services, and personalized and curriculum development, will serve a higher return. services for the world’s as vice president of educational activi- We have made advances toward this ties. In view of the soon-to-be-com- computing professionals.” goal. Recent examples include the pleted computing curriculum projects Distance Learning Campus, created in (http://www.computer.org/education/ association with KnowledgeNet, and cc2001/), including software engineer- the online bookshelf, powered by Second Vice President Richard A. ing and computer engineering, Murali Books 24x7. This new Society infra- Kemmerer, a professor and past chair intends to plan for more curriculum structure, which may take several years of the Department of Computer assessment and development. to fully implement, will lay the foun- Science at the University of California Michael R. Williams, professor dation for us to meet the challenges we at Santa Barbara, will chair the emeritus of computer science at the will face in the next decade. Chapter Activities Board. Under his University of Calgary and head cura- SP-5 will be presented for final leadership, in 2004 the Society will tor at the Computer History Museum approval at the February 2004 Board bring chapter members closer to each in Silicon Valley, will serve as vice pres- of Governors meeting, and, once other and narrow the gap between ident of the Publications Board. Mike approved, it will be ready for action. chapter leaders and the Society. will be instrumental in our ever- Oscar N. Garcia, a past president of increasing effort to build synergy in the 2004 APPOINTMENTS AND GOALS the Society, who is a professor and publications business with other IEEE I am privileged to introduce the founding dean of the college of engi- entities. 2004 Executive Committee, a group of neering at the University of North Yervant Zorian, vice president and dedicated and capable volunteer lead- Texas, will continue his faithful service chief scientist of Virage Logic and chief ers. These outstanding professionals as the secretary of the Board. technology advisor of LogicVision, will and scholars, well known in their indi- Rangachar Kasturi, the Hood continue his appointment as vice pres- vidual fields, share the Society’s vision Professor and chair of computer sci- ident of technical activities. Yervant and common goals to serve you and ence and engineering at the University will focus his energy to help discern the profession. of South Florida, has made significant and build technical communities while Joining me in 2004 are Past President achievements while serving as the vice strengthening the base of each existing Stephen L. Diamond, president and president for publications for the past technical committee and council. He CEO of Picosoft Inc., and President- three years. He will now head the also welcomes your input to help iden- Elect Gerald L. Engel, a professor of Finance Committee as the Society’s tify emerging technologies. computer science and engineering at treasurer. the University of Connecticut. Both will James W. Moore, a senior principal APPRECIATION AND EXPECTATION interact with me on a daily basis and engineer for the MITRE Corporation, I would like to especially thank Past advise me on many fronts. I feel will continue chairing the Standards President Willis K. King for his guid- extremely fortunate to serve as presi- Activities Board as a vice president. ance and encouragement during my
January 2004 11 President’s Message
on-the-job training as president-elect. served as the editor in chief of IEEE Willis shared with me many of his pro- Software. Since then, he has volun- found thoughts and wisdom, and he teered in many ad-hoc assignments was always willing to help me prepare that were critical to the Society’s oper- for my presidential term. ations. Most recently, Mark helped us Special thanks also are due to complete a new business model in the Guylaine M. Pollock, who just com- SP-5 draft plan. pleted her service as IEEE Division V Last but not least, I would like to Director. Key volunteers in the Society thank our staff. I have long been an know well that Guylaine used her advocate of volunteer/staff partner- insight and skill to build alliances in ship. My favorite example is how I the IEEE arena and to keep us engaged worked with Angela Burgess, now the in deliberations about many critical Society’s publisher. Angela and I issues. The Society benefited tremen- formed an editor-in-chief/managing dously from her efforts. editor partnership that transformed I would like to thank Deborah K. IEEE Software into a genuine tech- Scherrer, who directs NASA science nology-transfer forum. REACH and technology-based education and Bob Care, Violet Doan, Lynne public outreach programs at Stanford Harris, John Keaton, Anne Marie Kelly, University, for her leadership during Dick Price, and other staff have all her term as first vice president for edu- helped me in many different ways. HIGHER cational activities. Deborah conducted I look forward the opportunity to a review of EAB activities in an effort deepen such partnerships during the Advancing in the IEEE to focus on our educational program year. The leadership of David Hennage, Computer Society can priorities. the Society’s Executive Director, will be elevate your standing in I also recognize the contributions of instrumental to our success. the profession. Fiorenza C. Albert-Howard, an inde- pendent consultant, who just con- Application to Senior- cluded two terms as a member of the am looking forward to a productive grade membership Computer Society Board of Governors. 2004. We have a new vision. We recognizes I appreciate Fiorenza’s participation in I have assembled an outstanding team ✔ ten years or more many Society activities, including of volunteer leaders and staff. Most of professional audit, awards, chapters, conferences, importantly, we have more than 20,000 expertise and the Constitution and Bylaws Com- volunteers who carry out our mission mittee. throughout the year. I applaud those Nomination to Fellow- Doris L. Carver, associate vice chan- volunteers, for without them, our grade membership cellor of Research and Graduate Studies Society would not have advanced so recognizes and a professor of computer science at far. We need their continuing involve- ✔ exemplary Louisiana State University, will continue ment and support to grow our profes- I accomplishments to lead our flagship magazine, Com- sion and evolve the world. in computer puter, as editor in chief. James D. Isaak, engineering an assistant professor of information technology in the School of Business at Southern New Hampshire University, GIVE YOUR CAREER A BOOST and Gene Hoffnagle, research and tech- I nology strategist, IBM Centers for Advanced Studies, will represent the UPGRADE YOUR MEMBERSHIP Society at the IEEE as our IEEE division directors. In addition, I would like to thank a www.computer.org/ newly elected member of the Board of Carl K. Chang is chair of the Depart- Governors, Mark Christensen. Mark ment of Computer Science, Iowa State join/grades.htm joined the industrial advisory board University. Contact him at president@ that I assembled back in 1991 when I computer.org.
12 Computer A T R A N D O M
right places, a frame of the size and shape to support that musculature, and Design weight reduced to the bare minimum required by strength-of-materials con- siderations. Which means that the ani- mal’s bones are only just strong enough Fragility to get the job done under nominal con- ditions, thus placing it at greater risk Bob Colwell for breakage under any others. A faster horse is a more fragile horse. Another example comes from my personal experience of having built a ne of the things that makes guitar that happened not to implode engineering so interest- upon initial stringing. Acoustic guitars ing—besides the fact that are small masterpieces of mechanical they’ll pay you to do it—is engineering. They have a resonant cav- the interplay between so ity—the curvy body of the instrument— Omany contravening influences: fea- for amplifying the acoustical output of tures, performance, technology, risk, vibrating strings. These strings are cou- schedule, design team capability, and pled to the front surface of the resonant economics. You can find a perfect bal- cavity via a bridge, a strip of dense ance among all of these, and still fail, Tying conceptually material over which the strings bend because you designed something the separate functions sharply into the body itself, where pegs buying public happens not to want. hold them in place. The other end of For instance, if your startup pro- together increases each string is connected to a tuning poses to design the world’s best type- mechanism that adjusts its tension, and writer, the venture capital community overall complexity hence its resonant frequency/pitch. probably won’t beat a path to your and thus has a real An acoustic guitar with steel strings door. It’s not 1998 anymore. But if you places approximately 100 pounds of don’t find a workable balance that’s at price. force on the bridge plate—the piece of least as good as the one your competi- wood that is glued to the soundboard tors arrived at, your project is in big and has a slot to hold the bridge. trouble. Project-wide blue-screen-of- Because the string “breaks over” the death. Abort, error, no retry. (Does it design at one extreme and a fast, com- bridge, this force translates into a worry anyone but me that we all rec- plex, always-on-the-edge-of-disaster torque that tries to lift the bottom edge ognize these terms so quickly? This design at the other. One of the quali- of the bridge plate and push the string must be a bad thing.) ties that change as you move along that end into the soundboard. It is possible to achieve fragility line is something I call fragility. For cultural reasons that cello play- without complexity. All you have to do ers and violinists don’t understand, gui- is shave the design margins to the very HORSES AND GUITARS tars are made with mirror finishes, and edge of disaster so that the product is Consider the thoroughbred race- guitarists are generally extremely sen- barely achieving its design target. The horse. Over hundreds of years, these sitive about scratching them because odd thing is that in many ways this is horses have been bred to do two the mirror finish mercilously exposes also the point of maximum operating things: run a race course as fast as pos- all such flaws. This same finish lets you efficiency and performance, two of the sible, and make more horses that are see the effects of the string tension on most commercially important aspects as fast or faster than themselves. The the guitar’s soundboard. Between the of any design. This translates into an parallel between this sport and the soundhole and the bridge, a depression inexorable force propelling a design microprocessor business is unsettling. occurs in the wood of the soundboard team toward higher output at the The only differences are that the win- as the torque pushes it down; behind expense of design margin, a recipe for ning engineers don’t get roses, and the the bridge, the soundboard wood complexity. losers don’t often get turned into glue bulges out slightly. I believe there is a continuum of fea- and gelatin. Luthiers face the basic challenge of sible design points between a demon- But what qualities make a fast making the soundboard thin enough strably safe, solid, and possibly stodgy horse? Extreme musculature in the to sound good, but thick enough that
January 2004 13 At Random
it can withstand the string torque for upon to inspect a floor that was sag- tronics were really susceptible to any at least several decades of constant ging beyond its design specs. Upon damage and that that equipment was pull. Their solution is to make the inspection, the design engineer discov- high enough to be safe. Since most of soundboard a thin plane of wood, sup- ered that the contractor had been using the countryside in which Mason had ported by a system of braces glued to the floor as a temporary storage loca- to operate was hilly and heavily the underside. In acoustic steel string tion for extremely heavy travertine forested, his main rotor blades often guitars, at least two of these braces slabs, a load of 15 kN/m2 on a floor accumulated both bullet holes and begin at the edge of the soundboard built to code requiring 2 kN/m2. As green, leafy vegetation. He had learned and cross in an X pattern, with the Salvadori points out, had the floor col- to judge from the air how thick the intersection artfully placed at the point lapsed, the design engineer wouldn’t branches were so he would know of highest soundboard deflection just have been to blame. But had the floor whether rotor or tree branch would in front of the bridge. The braces are collapsed when the contractor was yield first in the event of a close shaved down to the point where they moving the travertine across it, he encounter. have only enough mass and strength to might have been at least partially cul- Flying machines place a premium on provide the necessary support. pable, even though the code “did not weight, but Mason gives the distinct When finishing the instrument, com- demand such a high live load.” impression that the Huey was designed mercial steel string guitar makers gen- to the limits of reliable operation first, erally apply a thin polymer finish. resulting in an aircraft that could rou- However, very high-end classical gui- There are no simple tinely outperform its specs in the areas tars are finished with French polish, a answers to where to that really mattered. In a very real manual-labor-intensive process that target a design on the sense, helicopters are fragile machines, builds up a high gloss with an extremely fragility scale. but within that design space, the Huey thin layer of lacquer; the finish looks designers got something right. The DC- great and the guitar sounds great 3 and the A-10 Warthog are other air- because there’s less mass on the sound- craft that have achieved legendary board, but the finish can be scratched The question of whether engineers status among their users for overall off with just a fingernail. should design strictly according to toughness and trustworthiness in a In other words, while inexpensive required specs or attempt to exceed clinch. instruments are built like battleships, those specs based on their own judg- There’s no simple formula for the best-sounding ones are, in a word, ment and morals is a tricky one. As a achieving a high-performance, com- fragile. project manager, I would expect that petitive product that will pass the test most design engineers would struggle of time like these examples. For BUILDINGS to hit their stated targets, without instance, in Searching for Safety Structural engineers and building allowing much time for anything (Transaction Publishers, 1988), Aaron architects must grapple with similar beyond that. It would also worry me Wildavsky points out how difficult tradeoffs. In Why Buildings Fall Down greatly to hear that some project engi- it can be when a litigious public col- (Norton, 1987), Mario Salvadori gives neers were adding to the product’s fea- lides with basic physics and the laws examples at both ends of the structural ture set without having run those of probability. fragility continuum. He relates how a features through the project manage- dome roof collapsed in 1960 in ment and feature planning processes. A problem with crashworthiness Bucharest, Romania, under a surpris- Such activities could easily result in cases is that juries focus on only one ingly small load just two years after a higher project risk. part of the automobile, and under new pipe-based method of supporting only one set of circumstances. Since the dome was tried. Salvadori attrib- HELICOPTERS, PLANES, CARS cars can strike or be struck from any utes the failure to a “tendency to In Chickenhawk (Viking Press, angle, at varying speeds, it is ex- design very light dome structures 1984), a harrowing account of his tremely difficult to design a car that [which] stemmed from the designers’ career as a helicopter pilot in the affords optimal protection in all virtuosity and competitiveness, but, Vietnam War, Robert Mason relates types of accidents. In one case, an more important, from economic con- many examples of having pushed his expert witness testified that injury siderations deriving from the high cost Bell Huey helicopter far beyond its was enhanced because a seat was of materials.…” design limits. For example, he once anchored and failed to give way. In a At the other end of the fragility scale, “landed” in a small river so that the different case, experts argued that Salvadori relates an incident in which rushing waters would clean the cargo failure to anchor a seat had led to a building design engineer was called deck off, knowing that only the elec- greater injuries.
14 Computer Automobile air bags that save many improvement hides too much detail. If nature, surprises unerringly make a lives but cause death or serious injury we look behind that curtain, we see beeline for the bad news camp, from in a small percentage of crashes are that over the same time period, the which they proceed to make the pro- another example of this phenomenon. sheer number of software applications ject leader pull out more hair. Spend There are no simple answers to where has increased by the same exponential the higher transistor budget in keeping to target a design on the fragility scale, amount, and the average delivered per- the integer and floating-point sides sep- other than this one: Don’t sacrifice formance varies widely from one appli- arated, and many benefits will accrue. everything to performance. Make all cation to another. When I was first learning to fix tele- such tradeoffs purposefully. visions—yes, they were black and white Exponential complexity back then—it bothered me that the hor- COMPUTERS It’s not just a matter of language, or izontal sweep and the high voltage cir- The average performance of com- coding style, or compiler differences, cuits were closely intertwined. I know puter systems has skyrocketed over the although they’re all important factors. why that was done—it’s easier to gen- past three decades. We architects get The root cause is that overall computer erate high voltage from an alternating more of the credit for this than we actu- system complexity has increased expo- source because a simple step-up trans- ally deserve. After all, the migration nentially, and complex systems behave former followed by rectification suffices. from the single-issue scalar processors in complex ways. Some applications In terms of understanding televisions of the 1970s to the out-of-order like caches; some like the L2 cache but in the first place, it’s easy to see that extremely deeply pipelined multi- won’t fit in the L1; some map onto the something must sweep the beam from threaded firebreathing “funny cars” of branch prediction scheme perfectly, top to bottom, something else from left today was achieved by a combination and a few are pessimal for it. to right, with some means of attract- of microarchitecture improvements and ing the beam toward the front of the process technology advances. set from the hot cathode that emits the There’s always a cost electrons. But armed only with that Architecture versus technology in tying conceptually understanding, and facing a TV with According to Modern Processor separate functions no raster, you might be misled into Design by John Paul Shen and Mikko together. thinking the set had a high-voltage H. Lipasti (McGraw-Hill, 2003), problem when really it was the hori- microarchitecture advances have zontal oscillator that had died. led from 0.1-instructions-per-cycle All of this leads to ever-widening There’s always a cost in tying con- machines in 1970 to 2-IPC machines variability in the performance ceptually separate functions together. in 2003, a 20-fold improvement. In extremes of applications on each new, Such an expedient increases the overall that same period, clock rates increased more complex processor. In other complexity and thus has a real price. from 0.1MHz to today’s 3GHz, a words, we’re inexorably moving 30,000-fold improvement. So who gets toward the more fragile end of the BAD BREEDING the credit for faster chips—the archi- scale at the system level. Reducing this to pithy T-shirt apho- tects or the process technologists? There are ways to spend higher tran- risms, I propose the following formu- This isn’t quite fair, of course. Had sistor budgets to move to the left, lation: the architects not provided microar- towards safety, on the fragility scale. chitectures capable of being run at Some microprocessors have been • Complexity breeds fragility. those fast clock rates, numbers like designed without integer multipliers, for • Fragility breeds surprises. 3GHz would not have been attained. example, because IMUL hardware is • Surprises are bad. And designing for fast clocks causes a large and power-hungry. After all, the decided loss in efficiency as measured instruction set architecture already Strive at all times to maintain a sense by instructions/clock. Taking that into requires the chip to translate integer to of the overall complexity of a given account, I still score process technol- floating point and back again. design, and place great importance on ogy as being at least two orders of I’ve never liked this idea. It may be minimizing it. magnitude more important than archi- faster than not having an IMUL, but Early in my career, I was riding in a tecture in terms of delivered perfor- not by much. Worse, it surprises pro- car with the chief architect of a micro- mance over that time period. grammers and compiler writers. processor that was then in its late It doesn’t really matter. These speed Always remember: Surprises during design stages, and he was holding his improvements have made today’s design aren’t evenly distributed into head in both hands, very depressed. computers useful for a wide range of good news and bad news categories. When I asked him what was wrong, he activities. But citing a 30,000-fold By some as yet undiscovered law of said, “I feel I have lost all control on
January 2004 15 At Random
this project. All of the intellectual on a slippery road or that are making engineering designs must avoid boundaries we placed on the various odd noises are similarly signalling their fragility whenever possible. The real units have gradually been compro- drivers. In each case, the physics of the key is keeping the design objectives in mised. The designers are freely bor- situation provides an operational mar- sharp focus. rowing time and space from each gin that can be of great value. If physics dictates that fragile guitars other, or sharing functionality for the We in the computer industry have an sound best, then a luthier can direct her sake of die area. Nobody knows how extra challenge in trying to design less efforts to that balance point, and she the chip is going to work any more. I’m fragile, more forgiving systems. Un- can educate her customers about the drowning in a sea of complexity, and I fortunately, computer-controlled sys- best ways to handle that system weak- can’t see how to get back out of it.” tems don’t naturally fail gracefully. One ness. In the end, it’s that luthier’s cus- Only through a Herculean project moment, the computer is monitoring tomers who will decide whether a management campaign and many sili- many different sensors and using that better-sounding but less forgiving fin- con respins was that project salvaged. information to efficiently control actu- ish is the right choice. Insidiously, only the core architects and ators and motors; the next moment, the In our world, if you must design a the project leaders could see the real computer has failed completely and is fragile system, then do it purposely, culprit: the cost of complexity. accomplishing precisely nothing, leav- carefully, and with the same attention One reason that Chuck Yeager was ing the chemical plant, or airplane to detail that your luthier counterpart a great test pilot was his innate sense engine, or elevator to run uncontrolled would have applied. Remember that of how close to the edge his aircraft toward maximum entropy. unless you actively manage it through- was. He could feel it and hear it, and Graceless failure, often with no out a design project, system fragility has under emotional pressure that would warning, is the hallmark of a computer a way of sneaking up on you. Eschew crush most of us, he could relate that system that was designed around per- complexity. Surprises are bad. sense to his intellectual understanding formance targets without regard to of the aerodynamics of flight and take system fragility. Bob Colwell was Intel’s chief IA32 the right action. Boats that are low in architect through the Pentium II, III, the water and responding sluggishly to he best-sounding guitars are the and 4 microprocessors. He is now an the helm give the same situational most fragile, but it’s clear from independent consultant. Contact him sense to an alert skipper. Cars that are T the other examples cited that at [email protected].
2004: January July looking outlook issue next-generation search February August ahead ad hoc networking sensor networks March September to hardware speculation robotics April October future computers and the aging software architectures technologies May November adaptive software and information assurance hardware December June next-generation personal internet security computing To submit an article for publication in Computer, see our author guidelines www.computer.org/computer/author.htm 1972•1988• 197232 & 16•1988 YEARS AGO
BATTLING TRASH. “Faced with moving a quarter million Editor’s Note: The more we know about yesterday, the tons of trash a year off District streets, the Department of better we will be able to deal with today. Computer Environmental Services, Washington, D.C., has enlisted a offers this bimonthly column providing excerpts from land pollution index and an IBM computer to help the city’s past issues to serve as a memory jogger for older mem- war on grime. Both the index and computer are part of bers and as a perspective creator for newer members. phase II of Operation Clean Sweep, the city-wide program designed to give District residents clean streets and, at the JANUARY 1972 same time, a new sense of community pride and spirit. PUBLIC CONCERNS. “Addressing an audience of computer “Department inspectors will match every street against a specialists, industry executives, management personnel, edu- series of photographs that set cleanliness standards. If a cators, and user industry representatives, Dr. Bruce Gilchrist, street doesn’t measure up, it’s reported to the Department AFIPS Executive Director, stated, ‘While 71% of those sur- so immediate action can be taken. The Department is cur- veyed believe life is better today because of the use of com- rently using an IBM System/360 Model 50 to help keep puters, there is major concern in a number of areas. These track of the 83 trucks that travel the 165 routes and stop at include automation and unemployment, increased depen- 135,000 trash pick-up points each week. dence on computers, dehumanization, the gathering of “The computer is also saving manpower for the District information on people by large organizations, and the pos- by cutting the time it used to take to re-configure a route by sible misuse of computerized information files.’” hand from 15 days to only one.” COMPUTER ON A CHIP. “Intel has introduced an integrated CPU complete with a 4-bit parallel adder, sixteen 4-bit reg- JANUARY 1988 isters, an accumulator and a push-down stack on one chip. ELECTRONIC PUBLISHING. “The impact of electronics and It’s one of a family of four new ICs which comprise the modern computer technology on the creation and publi- MCS-4 microcomputer system—the first system to bring cation of written information has just begun to be felt. One the power and flexibility of a dedicated general-purpose can safely predict the non-death of paper; the paperless computer at low cost in as few as two dual in-line packages. society is just not going to happen. In fact, expect just the “MCS-4 systems provide complete computing and con- reverse. The information glut will continue to expand. The trol functions for test systems, data terminals, billing new electronic media won’t replace paper, but augment it machines, measuring systems, numeric control systems and by providing more efficient and effective access to infor- process control systems.” mation. And when that information is found, a paper copy will be made.” BAD ART. “A California graphics design company is spon- soring a ‘Bad Art’ contest to call attention to what the firm perceives as a lack of design standards in the electronic pub- lishing industry. The ‘winning’ entries will be hung in the firm’s Hall of Shame. “Bruce Ryon, president of Design Access, said the con- test was inspired by the recent flood of ‘graphically atro- cious’ ads created with EP systems. “‘The more hapless efforts... of aesthetically impaired desktop publishing ... should be preserved for posterity, like TV bloopers, man-bites-dog headlines, or those early films of airplanes that never got off the ground,’ Ryon said.” COMPUTER ON A CHIP. “Motorola has completed evalu- ation sampling of the 68030 32-bit microprocessor, called the ‘030.’ The company announced availability of 16- and 20-MHz chips at the same time that it announced devel- opment of a 25-MHz version. “According to the company, the 030 provides twice the performance of the 32-bit 68020 while remaining fully com- patible. It is reputedly the first microprocessor to have on- Used with permission from Blinkenlights Archaeological Institute chip data and instruction caches, parallel (Harvard-style) Intel’s MCS-4, described in 1972 as the first system to offer a 4-bit architecture, and dual modes of address.” parallel adder, sixteen 4-bit registers, an accumulator, and a Editor: Neville Holmes; [email protected]. push-down stack on one chip.
January 2004 17 INDUSTRY TRENDS
to pay for expensive designs. Their processors for PCs and servers gener- Vendors Go to ally handle applications, such as Microsoft Word, that perform many different types of tasks and thus exe- cute many types of instructions on rel- Extreme Lengths atively small data sets. Baron explained, “Workloads for the chips made by major manufactur- ers tend to be instruction intensive. for New Chips Therefore, a key way to improve these chips is to increase the number of exe- Steven J. Vaughan-Nichols cuted instructions per clock cycle. The best way to increase the number of executed instructions per clock cycle is by increasing a chip’s frequency.” s demand has grown for Parallelization approaches don’t wireless communications and work with the types of applications multimedia devices—includ- that run on traditional CPUs, Baron A ing cameras, MP3 audio said, because the programs usually players, IEEE 802.11 wireless handle highly interdependent data. LAN base stations, and cellular With interdependent data, a chip must phones—so has the demand for spe- resolve one group of information cialized embedded chips that can han- before starting on another. dle these products’ need for data- In addition to higher frequencies, intensive, fast, power-efficient, reliable Baron said, the major chip makers performance. improve their products via smaller fea- Neither communications nor the of school want to take experimental ture sizes, more transistors, and com- rendering of multimedia files tolerates designs and try them in the field.” Also, plex architectures and design. chips that cause transmission delays he said, many young engineers prefer This entails fitting and intercon- and interruptions. At the same time, making new designs at a startup to necting more components within the many of these processors must be writing device drivers for a major semi- same amount of space, which requires energy efficient because they operate conductor firm. large investments in computer-archi- within battery-powered platforms. And it’s not just wireless and multi- tecture research and more costly semi- This makes traditional CPUs unsuit- media technologies that are driving the conductor-manufacturing processes. able in many cases. extreme-chip market. Mike Calise, pres- To meet the demand for a different ident of extreme chip maker Clear- Extreme embedded chips type of processor, vendors are begin- Speed Technology, said, “We’re seeing Extreme chip vendors, Baron said, ning to offer unconventional extreme an unbelievable number of compute- are often younger companies with less chips, explained Max Baron, principal intensive operations in the biotechnol- money and fewer employees. They typ- analyst at In-Stat/MDR, a market ogy and scientific fields.” For this ically design chips that perform only a research firm. market, ClearSpeed makes a 32-bit few types of functions. They thus exe- coprocessor for PCI-architecture Pen- cute a small range of instructions on DOING MORE WITH LESS tium class workstations. The chip runs applications with a large amount of Driving the development of extreme at only 200 MHz but performs at 25.6 data that isn’t interdependent. chips is the huge growth in wireless Gflops per processor because it uses 64 For example, extreme chips in a communications and the resulting on-chip floating-point processors. device that handles multiple wireless demand for millions of mobile chips, LAN connections would process sev- said In-Stat senior analyst Tom Traditional CPUs eral similar data streams that could be Halfhill. According to Baron, there are two handled independently. However, Halfhill noted, big bucks broad types of chips. The first is made In this case, improving performance are not the only motivation. “It’s also by established companies such as entails increasing the amount of the engineering challenge,” he ex- Fujitsu, IBM, Intel, Motorola, and data—not the number of instruc- plained. “Lots of engineers coming out Texas Instruments that have the money tions—processed per unit of time.
18 Computer The lack of data interdependency makes it easier to improve perfor- Instruction memory mance via parallelization, Baron explained. Increasing chip frequency could also increase performance, but it Instruction register would come at a higher price and at the Instruction decoder cost of greater power consumption and PC heat, all unacceptable for the devices that use extreme chips. Fully connected Therefore, extreme-chip vendors gen- network erally develop massively parallel proces- sors that include multiple reduced-
Register Hazard Pipeline detection instruction-set computer (RISC) cores, files controller Sequencing logic called processor engines. Because they forwarding control Bypass networks and improve performance via paralleliza- Fully connected tion, these chips can typically operate at network lower frequencies than traditional CPUs. In essence, the key to extreme Function Function Data memory design is to miniaturize multiple simple unit unit processors and place them on a chip. Peter Claydon, chief architect of Source: Silicon Hive picoChips Designs, an extreme-proces- sor design consultancy, said, “Our Figure 1. Silicon Hive’s ultra-long-instruction-word extreme chip eliminates traditional PC102 contains 344 processors on a processor elements such as bypass networking and forwarding control, pipeline control, single chip that runs at [only] 160 and hazard detection. The compiler handles these jobs instead. This reduces control over- MHz. These processors communicate head, moves complexity to the compiler, and adds flexibility otherwise limited by a hard- with each other over an onboard, inter- ware-fixed instruction set. nal bus. And multiple tasks can be run at once. For example, a single PC102 and when new functionality was approach in its customized chip designs [used in a cellular base station] can needed, they had to redesign the hard- for communications and digital-signal process the telecommunications data ware. “This was an expensive and processing. for hundreds of cellular users in paral- time-consuming process,” Halfhill According to the company’s chief lel.” noted. “Now, designers can make the architect, Dirk Helgemo, instruction processor configurable so it will be sets are designed into the chip, rather Creative approaches much easier to adapt to changing stan- than streamed onto it as is usually the Several techniques have enabled the dards such as the constantly evolving case. This lets MathStar chips use mul- design of useful extreme chips. 802.11 Wi-Fi security standards.” tiple external I/O to handle large Better design processes. Chip design However, configurability can come amounts of data and thus enables mas- was more difficult in the past, Halfhill at a price. sively parallel processing without mas- said, largely because designers had to Jeroen Leijten, chief architect of the sive internal data buses. manually map out chip designs. Today, Philips subsidiary Silicon Hive, said that MathStar’s Field Programmable there are two main easy-to-use hard- for its configurable, ultra-long-instruc- Object Array uses data paths that are ware-design languages that make the tion-word Avispa+ processor to run loosely coupled between multiple chip-design process easier and faster properly, “all data pipeline staging must processors to form a multiple-instruc- by letting developers design a chip be explicitly scheduled in the compiler.” tion-multiple-data machine on a chip, as a software model: the Very High- Thus, the vendor must program not enabling it to handle data-intensive Speed Integrated Circuits Hardware only the chip’s general operations but applications. Description Language and Verilog. also how the chip will deal with a spe- Once compiled, developers can manu- cific program. This yields a fast, highly Challenges to doing ally tweak chip designs generated via customizable processor, as Figure 1 more with less these languages to make them more shows, but generates more work and Extreme processors face several efficient. expense for the vendor. technical and marketplace challenges. Configurable processors. In the past, Creative parallel processing. Math- For example, extreme chips may not vendors put functionality in hardware, Star uses a creative parallel-processing appeal to some potential users who
January 2004 19 Industry Trends
don’t understand their unconventional processors,” Halfhill said, “but some many different types of demands. architectures or don’t want to deal applications may require multiple RISC Thus, he concluded, there won’t be the with new vendors. chips and DSPs to match the perfor- standardization in the extreme-chip Development tools. Most extreme mance of a single unconventional market that you see in the PC or server processors require special develop- processor.” However, using multiple chip market. ment tools. Baron explained, “You chips like this would cost too much, Meanwhile, stated Carnegie Mellon simply can’t use a QNX, Wind River, consume too much power, and take up University Assistant Professor Seth or Gnu C because their compilers too much space for many host devices. Copen Goldstein, demand for high- don’t [work with] the new architec- There are also application-specific performance, low-energy consumption, tures.” Many companies thus modify integrated circuits and systems on chip, and other capabilities that gave birth to Gnu C because it’s open source and which have high performance and are extreme chips will continue, ensuring already well understood, he noted. easier to program than extreme chips the technology’s ongoing popularity. Design issues. Extreme designs with but require a long time to develop and However, added Baron, extreme massively parallel processors still lack thus may be obsolete by the time they designs will continue to face competi- efficient compiling and efficient are finished. Field-programmable gate tion from traditional established chips. designs for communication between arrays are flexible but expensive. In the end, he said, the marketplace the on-board chips, according to Industry observers say there are will decide which designs will flourish Baron. Designers now must both cre- enough different uses and enough and which will fall by the wayside. ate the chip and try to work out the demand for these types of chips, as well best way for the processor engines to as for extreme chips, to provide a mar- Steven J. Vaughan-Nichols is a free- network with each other, which entails ket for each. lance technology writer based in additional time and cost. Arden, North Carolina. Contact him Marketplace alternatives. Users at [email protected]. who want to gain some of extreme he embedded market, Halfhill chips’ benefits can turn to alternative noted, has a huge range of appli- types of hardware in some cases. T cations, as disparate as rocket Editor: Lee Garber, Computer, 10662 Los “High-performance RISC processors guidance systems, pagers, and antilock Vaqueros Circle, PO Box 3014, Los Alamitos, and DSPs [digital signal processors] are brakes. He said there will have to be CA 90720-1314; [email protected] easier to program than [extreme] multiple architectures to manage these SET wireless networks gigabit Ethernet INDUSTRY enhanced parallel ports STANDARDS FireWire 802.11 token rings IEEE Computer Society members work together to define standards like IEEE 802, 1003, 1394, 1284, and many more.
HELP SHAPE FUTURE TECHNOLOGIES • JOIN AN IEEE COMPUTER SOCIETY STANDARDS WORKING GROUP AT computer.org/standards/
20 Computer CALL FOR PAPERS The 1st IEEE International Conference on Mobile Ad hoc and Sensor Systems
(MASS-2004) GENERAL CHAIR Dharma P. Agrawal October 25-27, 2004, Fort Lauderdale, Florida, USA University of Cincinnati http://www.ececs.uc.edu/~cdmc/mass PROGRAM CO-CHAIRS Mario Gerla University of California, Los Angles ® IEEE Jie Wu Florida Atlantic University Sponsored by PROGRAM VICE CHAIRS IEEE TC on Distributed Processing and IEEE TC on Simulation Marco Conti (MAC layer and lower layers) CNR-IIT, Italy Recent Research on wireless networks and mobile computing has concentrated on single-hop networks, where network nodes communicating directly to a fixed Prasant Mohapatra (Network layer) infrastructure such as cellular or satellite systems. While in Ad hoc and sensor University of California, Davis networks multi-hop scenarios such as conference, hospital, battlefield, rescue and Ivan Stojmonvic (Algorithms and Topology) monitoring scenarios are covered, where network nodes communicate via other University of Ottawa, Canada network nodes. This conference covers the area of ad hoc and sensor systems, from physical issues to applications aspects. Yu-Chee Tseng (Applications) National Chiao Tung University, Taiwan Possible topics of interest include, but are not limited to:
TUTORIAL CHAIR - MAC protocols for ad hoc and sensor networks (801.11, Bluetooth, etc) Stephan Olariu - Routing protocols (unicast, multicast, broadcast, geocast, etc) Old Dominion University - Power-aware and energy-efficient design - Directional and smart antennas for ad hoc networks PUBLICITY CHAIRS Jean-Pierre Hubaux - QoS, security, and fault-tolerant issues EPFL, Switzerland - Location and mobility management - Topology construction and coverage maintenance Kia Makki / Chi Zhou - Distributed Algorithms (self-configuration, scheduling, synchronization, etc) Florida International University - Data gathering, fusion, and dissemination
Tom La Porta - Operating system and middleware support Pennsylvania State University - Complexity analysis of algorithms - Analytical, mobility, and validation models LOCAL ARRANGEMENT CHAIR - Applications of ad hoc and sensor networks Mihaela Cardei Florida Atlantic University Paper submission
FINANCE CHAIR All submissions will be handled electronically. Authors should prepare a PDF of their Anup Kumar full papers. Papers must not exceed 10 single-spaced and two-column pages using at University of Louisville least 11 point size type on 8.5 * 11 inches pages. Electronic submission instructions will be published on the website http://www.ececs.uc.edu/~cdmc/mass together PUBLICATION CHAIR Xiuzhen Cheng with style files. George Washington University Tutorials REGISTRATION CHAIR Proposals for half and full day tutorials are solicited. The selection criteria include the Hong-Yi Wu expertise and experience of the instructors and relevance to the central themes of the University of Louisiana conference. Proposals of at most 4 pages, including 1-page biographical sketch, INDUSTRY LIAISON should be submitted to the Tutorial Chair, Stephan Olariu ([email protected]), by Bob O'Dea April 30, 2004. Motorola Important dates IEEE TC CHAIRS Chita Das (Distributed Pro cessing) Manuscript Submission due: April 15, 2004 Anup Kumar (Simulation) Acceptance Notification due: July 20, 2004 Final Manuscript Due: August 30, 2004 TECHNOLOGY NEWS
SECURITY THREATS Clients, not network administrators, Is Peer-to-Peer control P2P networking, which is described in the sidebar “The Ins and Outs of Peer-to-Peer Networks.” Individual users are responsible for Secure Enough for securing their files and directories. P2P thus circumvents many types of cor- porate security, including controls over who is allowed to enter workspaces Corporate Use? and which system resources visitors have access to. George Lawton Companies that run or are consid- ering running P2P programs are con- cerned that users could accidentally share important corporate data. Jim Lowrey, a security specialist at En- eer-to-peer technology is deavors Technology, said another being used more frequently major security challenge is controlling for corporate activities. P2P the incoming P2P traffic that can enter P enables work teams, often a company’s network. Users are con- geographically dispersed, to cerned that P2P applications could let share information and coordinate pro- viruses or hackers enter systems. jects online, noted Nimisha Asthagiri, “One of the newer concerns is law- development manager for security suits occurring from [a company] hav- projects at P2P software vendor ing copyrighted material on its net- Groove Networks. work and servers, facilitating the trad- In a February 2003 study, market ing of this material. Companies are get- research firm Frost & Sullivan esti- ting [hit] with million-dollar settle- mated that revenue from the sale of Groove Workspace, and Joltid’s Peer ments because they have not prohib- corporate P2P systems will increase Enabler. And if vendors agree on a ited the use of these materials,” from $42.8 million this year to $4.53 solid, standards-based, widely sup- explained Frost & Sullivan security billion by 2007. As Figure 1 shows, ported infrastructure that provides analyst Jason Wright. Frost & Sullivan also predicted that the interoperability and platform indepen- Kevin Rowney, chief technology offi- number of enterprise P2P clients in the dence, corporate P2P adoption could cer of security vendor Vontu, said, US alone will rise from about 100,000 increase even more. “Most of the corporations we deal this year to about 1.8 million in 2007. However, now that P2P is being used with would prefer to see P2P use Within a company, explained Rice for more important functions, security stopped entirely on their networks. University Assistant Professor Dan has become a critical issue. Thus, com- They frequently view P2P systems as Wallach, “P2P technologies might be panies such as BitTorrent, Intel, Joltid, exposing companies to an excessive used for instant messaging, file storage, Microsoft, and Sun are creating plat- amount of risk.” remote backup, or a number of other forms that simplify the development applications.” and deployment of secure P2P applica- Sharing sensitive information Many large vendors are beginning to tions. Vendors such as Applied Meta, A critical vulnerability is that most integrate P2P applications with the core Entropia, Groove, Popular Power, and P2P applications can turn a computer business programs they sell, such as United Devices have integrated stan- into a network file server. This their customer-relationship manage- dardized security approaches into their increases the chances of improper shar- ment systems. And vendors such as P2P products. ing of intellectual property, confiden- IBM, Intel, Microsoft, and Sun Micro- These developments lend credibility tial corporate information, and other systems are integrating P2P functional- to P2P’s suitability for corporate use, sensitive documents. ity into other products. as well as for sharing music files, said A recent US Congressional study, Moreover, there is a growing num- Jarad Carleton, leader of Frost & “File-Sharing Programs and Peer-to- ber of stand-alone corporate P2P appli- Sullivan’s Internet infrastructure pro- Peer Networks: Privacy and Security cations, including Avaki Data Grid, gram. Risks,” found that many users of con-
22 Computer sumer P2P applications accidentally made highly personal information— 2,000 such as tax returns, financial docu- 1,800 ments, medical records, and attorney- client communications—available to 1,600 other users. Similar problems could 1,400 occur with corporate P2P networks. In addition, P2P software bugs could 1,200 cause systems to accidentally make 1,000 sensitive information available. “A P2P client is also a server, in that it will 800 accept connections from other ma- Units (thousands) 600 chines on the network,” explained Rice University’s Wallach. “Like any 400 server, it needs to be robust against 200 buffer overflows and the like. Any vul- nerabilities in the P2P client would, like 0 vulnerabilities in a Web server, allow 2000 2001 2002 2003 2004 2005 2006 2007 Source: Frost & Sullivan an attacker to break in and take con- trol of the computer.” According to Wallach, the majority Figure 1. Frost & Sullivan, a market research firm, predicts that the number of corporate of P2P systems don’t include encryp- peer-to-peer clients in the US will increase steadily until 2005, when it will start rising tion and thus could let eavesdroppers more quickly. with sniffers or similar devices find and then view material sent by unsuspect- ing users. The Ins and Outs of Peer-to-Peer Networks Attackers could also steal private Peer-to-peer software lets users access files on other computers by creating information or even activity log files an interface to files that other users have made available via the P2P network. from a client by using a P2P applica- In some cases, systems implement P2P communications by giving each node tion to scan for an accessible directory both server and client capabilities. Users start by downloading and executing on a victim’s system or by exploiting a P2P networking program. After launching the program, users can enter the system vulnerabilities. IP addresses of other computers on the P2P network with whom they want to As is the case with other applications, communicate. the most difficult type of P2P-based Users then decide which of their files they want to share and which they data theft for organizations to prevent want to password-protect. When they want to communicate with another net- involves authorized users misappropri- work member, their computer establishes a connection online with the mem- ating or being careless with data. ber via the appropriate IP address. Usability problems Most consumer and many corporate P2P applications communicate as HTTP traffic on TCP/IP port 80, making communications from outside an Network administrators can set up organization difficult to stop with standard firewalls, which generally leave P2P applications with varying degrees the port open to give users access to the Web. of centralized control. However, users Sender-anonymizer and encryption technologies make it even harder for frequently maintain control of manag- firewalls and other technologies to analyze P2P messages and determine ing their own file directories and the whether to stop them. permission process for accessing them. For P2P applications to be secure, users must know how to configure them properly. For example, they must either correctly or in compliance with software (as well as system-clogging, ensure that they set up only the proper company policies can create serious activity-tracking spyware) as part of directories for sharing with other P2P security problems. files they access. They could also quickly users and that all data is identified as spread malware they accidentally ac- to whether it can be shared and if so, Spreading malicious code quire from e-mail or other sources. with which groups or individuals. P2P users could accidentally down- Consumer P2P systems have already The failure to perform these tasks load viruses, worms, or other malicious seen proof-of-concept worm attacks
January 2004 23 Technology News
such as VBS_GNUTELWORM and limit the access that other parties have Although effective, PKI can be com- W32/Gnuman. There is also concern to system resources. This requires plex, expensive, and time-consuming that P2P could spread malware in part authorization, in which a system defines to implement. because of software-design flaws. a user’s or group’s ability to access spe- Directory services. Companies could Sharon Ruckman, senior director at cific directories or sets of data. also use their existing enterprise direc- antivirus vendor Symantec’s Security tory services, which list the resources Response center, said its biannual Authentication available within a network, to identify Internet Threat Report found that of Because basic P2P applications lack users, find their public keys, and oth- the most prevalent 50 viruses and their own security, including identity erwise provide some P2P authentica- worms, the number that could attack authentication for those involved in tion. Companies such as Endeavors P2P systems increased from four in P2P communications, users can’t and Groove plan to take this approach 2002 to 19 this year. assume that other parties are who they with their products. say they are, explained Kevin Beaver, Users could accomplish directory- External attacks chief technology officer of security ven- based authentication by various means, There is considerable concern about dor Principle Logic. including the lightweight directory P2P systems opening a network to access protocol, which is a set of pro- external attacks. As is the case with Web tocols for accessing information direc- servers or e-mail servers, P2P software Vendors are looking at tories. can have bugs that hackers can exploit various ways to provide A directory usually covers a single by creating problems such as buffer security for P2P systems. administrative domain, generally rep- overflows, noted Rice University’s resenting a single organization or Wallach. department, and thus would help According to Frost & Sullivan’s Digital signatures can help authen- authenticate P2P users from that Wright, “If you are running a P2P app ticate P2P communications. The sig- domain but not from other domains. at home and using a [virtual private natures are digital codes that are network] into the [company’s] net- attached to an electronically transmit- Encryption work, you are essentially providing a ted message and that uniquely identify Encryption can play several roles in pathway for anyone who wants to scan the sender. P2P applications. For example, it can you as a host. There are ways to PKI. Peers can identify each other by protect a particular transmission be- manipulate that situation to gain directly exchanging credentials, such tween peers on an unsecured network. access to a corporate network through as digital certificates. Administrators can also use encryp- a VPN tunnel.” They can also use public-key infra- tion to create a VPN for communica- structure (PKI) technology, in which a tions over unsecured networks. In PROPOSED SOLUTIONS trusted third-party authority issues dig- addition, by using keys unique to spe- Vendors are looking at various ways ital certificates. A user who wants to cific users, encryption can support the to provide security for P2P systems. send an encrypted message applies for authentication process. For example, in its P2P products, a digital certificate from a certificate A few smaller companies, such as Endeavors uses WebDAV (Web dis- authority. The authority issues an Endeavors, have developed corporate tributed authoring and versioning), an encrypted digital certificate containing P2P systems that use the encryption- Internet Engineering Task Force stan- the applicant’s public encryption key based secure sockets layer (SSL) tech- dard that provides HTTP extensions and other identifying information. nology, which provides security for to facilitate collaborative editing and The message recipient uses the Internet messaging. Other firms are file management between users work- authority’s own, readily available pub- working with Pretty Good Privacy ing together over the Internet. lic key to decode the certificate encryption, used extensively with e- A new security model would peri- attached to the sender’s message. The mail. Other options include crypto- odically redirect P2P data from the recipient verifies that the authority graphic technologies such as the RSA client, where the user keeps it, to a cen- actually issued the certificate and then public-key encryption system with tral server for auditing, Groove’s gets the sender’s public key and identi- 2,048-bit keys and the 192-bit-key ver- Asthagiri said. This auditing could fying information from the certificate. sion of the Advanced Encryption identify problems that have already With this information, the recipient Standard algorithm. occurred and security trends that need can send an encrypted reply that only to be watched. the original sender can decrypt with a Intel’s Peer-to-Peer In P2P communications, as in many private key, which corresponds math- Trusted Library other types of transactions, users must ematically to the public key. Intel has a free, open source devel-
24 Computer opers’ library that companies can This could help the firewalls better prise, the future for this technology download to add security to P2P sys- detect and block sensitive P2P content could be very bright,” he added. “On tems they develop. The Peer-to-Peer being improperly sent outside a cor- the other hand, if vendors consistently Trusted Library (http://sourceforge.net/ porate network. fail to standardize around a platform projects/ptptl) supports digital certifi- Unlike today’s faster technologies, that addresses these concerns, network cates, peer authentication, secure stor- slower firewalls couldn’t handle the administrators will work to stop these age, encryption, and digital signatures. overhead of such extensive inspections. deployments.” The library is built on the open source OpenSSL toolkit (www.openssl.org/). George Lawton is a freelance technol- aid Vontu’s Rowney, “The enor- ogy writer based in Brisbane, Califor- Application-layer firewalls mous power of [P2P] systems to nia. Contact him at glawton@glawton. Traditional firewalls evaluate the S enhance the creative exchange of com. header information on packets to information, ideas, and content can’t determine whether to admit them to a easily be brushed aside by security con- network. Frost & Sullivan’s Wright cerns. Over time, companies will find said a new generation of application- productive ways to conduct business Editor: Lee Garber, Computer, 10662 Los layer firewalls, from vendors such as using these technologies.” Vaqueros Circle, PO Box 3014, Los Alamitos, Check Point Software Technologies “If vendors of P2P platforms com- CA 90720-1314; [email protected] and Sanctum, use deep packet inspec- mit themselves to addressing the rea- tion to examine a message’s contents. sonable security concerns of the enter- GET CERTIFIED 2004 Test Windows: 1 April—30 June and 1 September—30 October Applications now available! CERTIFIED SOFTWARE DEVELOPMENT PROFESSIONAL PROGRAM
Doing Software Right Demonstrate your level of ability in relation to your peers Measure your professional knowledge and competence Certification through the CSDP Program differentiates between you and other software developers. Although the field offers many kinds of credentials, the CSDP is the only one developed in close collaboration with software engineering professionals.
“The exam is valuable to me for two reasons: One, it validates my knowledge in various areas of expertise within the software field, without regard to specific knowledge of tools or commercial products... Two, my participation, along with others, in the exam and in continuing education sends a message that software development is a professional pursuit requiring advanced education and/or experience, and all the other require- ments the IEEE Computer Society has established. I also believe in living by the Software Engineering code of ethics endorsed by the Computer Society. All of this will help to improve the overall quality of the products and services we provide to our customers...” — Karen Thurston, Base Two Solutions
Visit the CSDP web site at http://computer.org/certification or contact [email protected]
January 2004 25 NEWS BRIEFS
E-Mail Security Gets Personal
US company has developed private key based on the public one. In to customers,” said Ahmad, because and released a simplified, less addition to providing public-key tech- the technology is expensive and diffi- expensive e-mail security sys- niques, Ahmad said, Voltage’s tech- cult to use. tem that derives its public nology also offers strong elliptic-curve By automatically using an e-mail A encryption key from a mes- cryptography. address to create keys that enable sage sender’s own e-mail address. Voltage says its approach automates trusted, secure communications, Voltage Security’s SecureMail soft- and simplifies the authentication Voltage’s technology eliminates PKI’s ware suite, which works with desktops process, reversing the trend of increas- requirement that communicating par- or laptops, transparently executes the ing complexity in technology that ties always enroll with and receive cer- cryptography, making it as simple as establishes trust between two commu- tificates from a third party attesting to sending unencrypted e-mail, said com- nicating parties. This could attract their identities. pany spokesperson Wasim Ahmad. users who want more secure e-mail but With PKI, senders who have SecureMail’s server-based software are concerned about the amount of received certificates must find a recip- uses one algorithm to convert an e-mail work necessary to implement public- ient’s public key in a directory and use address into a large number that func- key infrastructure technology, the tra- that key to scramble a message. tions as a public encryption key and ditional way to encrypt e-mail. Recipients then must unscramble it another algorithm to create the user’s “You can’t roll out traditional PKI using a private key that is mathemati- cally related to the public one. Thus, PKI entails a complex, costly process that requires overhead such as certifi- Company cate management. With Voltage, only the sender needs Internal network the software in advance. Recipients running various e-mail applications Voltage SecurePolicy Suite can click on a link to download free Voltage software that automatically decodes the message and lets them send an encrypted response. They can Enterprise Management console Internet also use a no-download reader that lets authentication them read the mail in a browser but
Public not send an encrypted response. parameters Because of the direct communica- (hosted on tion between senders and recipients, Web server) Enrollment Key the Voltage system further reduces server server complexity and potential management Firewall Firewall problems by not requiring a dedicated server to handle the secure transmis- Desktop sions. To succeed, Voltage must determine Source: Voltage Security how to get enough users to work with the technology for it to be effective, explained Phebe Waterfield, analyst for Voltage Security has released an e-mail security system that automatically authenticates the Yankee Group, a market research users and derives its cryptographic keys for encoding messages from a user’s own e-mail firm. And when widely used, she address. This is simpler than public-key-infrastructure technology, a traditional but com- added, the approach must prove to be plex way to secure e-mail. Voltage users who work with the system for the first time as transparent as Voltage claims. The authenticate themselves to the enrollment server. Once authenticated, they receive keys system has yet to be rigorously and from the key server. The public parameters contain all the information an organization independently tested, she noted. needs to create keys for any user. —Linda Dailey Paulson
26 Computer Ceiling for Controversial US H-1B Visas Drops he US has reduced the number of time when domestic unemployment is L-1 visas, which permit intracompany people admissible under its con- a politically sensitive issue. transfers of foreign managers and T troversial H-1B visa program, The AILA estimates 900,000 H-1B workers with specialized skills. which lets skilled foreign workers— employees, 35 to 45 percent from Some businesses say that they plan many in the computer technology India, currently work in the US. to cope with lower H-1B visa caps by field—take jobs in the country for up Steadman said US technology firms outsourcing work overseas. to six years. are currently finding some of the over- The US Congress refused to change seas employees they want by using —Linda Dailey Paulson a policy, called for by the American Competitiveness in the 21st Century Act (AC21), that decreased the annual Technique Speeds Data Transmission in Computers visa ceiling beginning this fiscal year to 65,000. The cap the past three years Sun Microsystems researchers have discovered a way to transmit data inside was 195,000. a computer 60 to 100 times faster than with traditional techniques by letting Paul L. Zulkie, president-elect of the information pass directly between chips placed next to one another. The American Immigration Lawyers’ approach would also reduce energy consumption. Association (AILA), which supports The Sun researchers say they sent data at 21.6 Gbits per second between chips higher H-1B caps, predicts the US will using a pared-down version of their new technology. Principal investigator award all 65,000 visas by March 2004. Robert J. Drost said this prototype was a proof of concept that used chips with He said this would leave many com- 0.35-micron feature sizes rather than chips using today’s smaller feature sizes. panies without the qualified employees Speeds could be even greater using current chip technology, he noted. they need until the next fiscal year Traditionally, data passes between elements in a computer via wiring. With starts on 1 October. Sun’s approach, chips sit next to each other and transmit data between con- The H-1B visa program has sparked nection points, which include tiny transmitters and receivers, via capacitive cou- ongoing controversy. Many technol- pling. When chips are brought into alignment, explained Drost, the capacitance ogy companies say they hire US work- between them carries the signal across the short open space between the con- ers when possible but need higher visa nection points with very little of the latency that occurs when signals have to pass caps to fill important positions requir- through wires. ing skills for which qualified domestic A system’s total throughput would increase with more interchip connections. workers aren’t readily available. Thus, Drost said, when Sun’s technique is used fully in computers, data rates H-1B opponents, on the other hand, could exceed a trillion bits per second. have assailed the program for taking Currently, the 25-micron wiring that takes data into and out of chips is sol- away scarce technology jobs from US dered to processor edges at ball-bond pads with pitches 250 microns wide. With workers and giving them to workers Sun’s approach, connection points would be only 50 microns wide. Thus, there from other countries who, they say, is room for more data connections, which leads to higher transfer rates. often accept lower salaries. Meanwhile, because the system only has to push signals across a small open- Zulkie said AC21’s temporarily ing between connection points, it uses less power. higher visa caps failed to provide a Sun’s approach would also improve performance by letting computer design- long-term solution to US employers’ ers pack chips into a system far more densely than is possible today. need to hire foreign professionals. However, because the chips would be in close proximity, there is concern that However, president John Steadman of the approach could generate excessive heat and that the connection points could the IEEE-USA, which opposes high H- create interference. Some also question whether the technique can be imple- 1B, said the unemployment rate for US mented effectively on an assembly line. IT professionals is high enough that “It’s an extremely difficult problem. [Sun’s approach] is unique and holds a many domestic engineers are available great deal of promise,” said Christopher Willard, research vice president with to hire. market research firm IDC. Harris Miller, president of the Although Sun hasn’t fully tested its technique, Drost said, the US Defense Information Technology Association Advanced Research Projects Agency has expressed interest in it. Meanwhile, the of America, an H-1B supporter, said company says it wants to find commercial implementations as soon as possible. the US Congress didn’t want to con- —Linda Dailey Paulson sider approving a higher ceiling at a
January 2004 27 News Briefs
Viewing the World through Interactive Panoramic Images
hotographs limit viewers to see- For one thing, high-resolution digital noted, fast Internet connections ing the image captured in the cameras have minimized the work improve the viewing process by quickly P viewfinder, no matter how inter- needed to build panoramas from indi- delivering detailed, full-screen images. esting the surrounding areas might be. vidual photos by producing images that Other photographers are experi- However, interactive panoramic images don’t have to be digitized first. Also, menting with ways to expand the are beginning to solve this problem and Goetze said, getting images on a Web viewer’s experience by providing create a new art form. server is easier and quicker than in the video-like transitions to link separate The increasingly popular panoramic past. It’s now possible to produce a but related sets of panoramic images. images let viewers move their cursor working panorama perhaps 5 minutes Also, companies such as Zoomify across a photograph and get a 360- after shooting it, he explained, although are developing software that can zoom degree view of a scene beyond the small more work improves the final product. in on panoramic images and stream image visible in a browser window. Creating a panorama also requires additional detail into a scene as the Panoramas used to be made primar- stitching multiple images together and viewer moves “closer.” ily by hobbyists who took a series of editing the image to make sure the A few businesses are exploring aug- photos across a scene and then used a stitching was done precisely. Stitching mented panoramas, which are computer to assemble them horizon- applications—such as Apple’s Quick- designed to make a scene more inter- tally. They displayed the finished prod- Time VR Authoring Studio, Pano- esting by adding animated characters uct in a browser window in which Tools, and RealViz Stitcher—analyze or streaming Webcasts of people. viewers could move from side to side image pairs and work with common However, this approach is not com- to see the entire view. features to match them properly. mon because it is time-consuming. Panoramic-image creation now Some applications let designers cre- Fred Ritchin, director of the online appears to be expanding for several ate spherical panoramas in which a magazine Pixel Press and an associate reasons, said Web designer and pho- viewer can look up and down as well professor of photography and commu- tographer Erik Goetze. as side to side. Meanwhile, Goetze nications at New York University’s Tisch School of the Arts, said designers haven’t begun to explore panoramic Web Sites that Offer Panoramic Images images’ full potential. For example, he said, they could be The following Web sites offer panoramic images, some of which require used with text and other types of con- plug-ins to view: tent to create a more interesting Web • The Picture Project’s 360degrees site is a social-documentary project on experience. the US criminal justice system: www.360degrees.org. • Australian commercial photographer Peter Murphy started a Weblog showcasing panoramas: www.mediavr.com/blog. He also has panoramic images at www.mediavr.com/bronte1.htm and http://culture.com.au/ Linda Dailey Paulson is a freelance virtual/. technology writer based in Ventura, • 360 VR Studio, with New Jersey commercial panoramic photographer California. Contact her at ldpaulson@ Jook Leung, offers panoramic images shot in the New York City area: yahoo.com. www.360vr.com. • Danish commercial photographer Hans Nyberg maintains a Web site with links to other sites containing 30,000 panoramas: www.panoramas.dk. • Belgian photojournalist Tito Dupret has panoramas of places such as Beijing’s Forbidden City and Cambodia’s Angkor Wat on the World Heritage Tour Web site: www.whtour.net. Editor: Lee Garber, Computer, 10662 Los • An augmented panoramic image can be found at www.throbbing. Vaqueros Circle, PO Box 3014, Los Alamitos, com/bikerace. CA 90720-1314; [email protected]
28 Computer PURPOSE The IEEE Computer Society is the PUBLICATIONS AND ACTIVITIES world’s largest association of computing Computer. An authoritative, easy-to-read professionals, and is the leading provider of magazine containing tutorial and in-depth technical information in the field. articles on topics across the computer field, MEMBERSHIP Members receive the plus news, conferences, calendar, industry monthly magazine Computer, discounts, trends, and product reviews. and opportunities to serve (all activities Periodicals. The society publishes 12 are led by volunteer members). Member- magazines and 10 research transactions. ship is open to all IEEE members, affiliate AVAILABLE INFORMATION To obtain more information on any of the Refer to membership application or request society members, and others interested in information as noted at left. the computer field. following, contact the Publications Office: • Membership applications Conference Proceedings, Tutorial COMPUTER SOCIETY WEB SITE Texts, Standards Documents. The IEEE Computer Society’s Web site, at • Publications catalog The Computer Society Press publishes www.computer.org, offers information • Draft standards and order forms more than 160 titles every year. and samples from the society’s publications and conferences, as well as a broad range • Technical committee list Standards Working Groups. More than 200 groups produce IEEE standards of information about technical committees, • Technical committee application standards, student activities, and more. used throughout the industrial world. • Chapter start-up procedures OMBUDSMAN Members experiencing Technical Committees. Thirty TCs pub- problems—magazine delivery, member- • Student scholarship information lish newsletters, provide interaction with ship status, or unresolved complaints— • Volunteer leaders/staff directory peers in specialty areas, and directly influ- may write to the ombudsman at the • IEEE senior member grade applica- ence standards, conferences, and education. Publications Office or send an e-mail tion (requires 10 years practice and sig- Conferences/Education. The society to [email protected]. nificant performance in five of those 10) holds about 100 conferences each year CHAPTERS Regular and student chapters To check membership status or report a and sponsors many educational activities, worldwide provide the opportunity to change of address, call the IEEE toll-free including computing science accreditation. interact with colleagues, hear technical number, +1 800 678 4333. Direct all other experts, and serve the local professional Computer Society-related questions to the community. Publications Office. EXECUTIVE COMMITTEE President: VP, Educational Activities: VP, Standards Activities: 2003–2004 IEEE Division VIII CARL K. CHANG* MURALI VARANASI* JAMES W. MOORE† Director: Computer Science Dept. VP, Electronic Products and VP, Technical Activities: JAMES D. ISAAK† Iowa State University Services: YERVANT ZORIAN† 2004 IEEE Division VIII Director- Ames, IA 50011-1040 LOWELL G. JOHNSON (1ST VP)* Secretary: Elect: STEPHEN L. DIAMOND* Phone: +1 515 294 4377 VP, Conferences and Tutorials: OSCAR N. GARCIA* Computer Editor in Chief: Fax: +1 515 294 0258 CHRISTINA SCHOBER* Treasurer: DORIS L. CARVER† [email protected] VP, Chapters Activities: RANGACHAR KASTURI† Executive Director: President-Elect: RICHARD A. KEMMERER (2ND VP)† 2003–2004 IEEE Division V DAVID W. HENNAGE† GERALD L. ENGEL* VP, Publications: Director: * voting member of the Board of Governors Past President: MICHAEL R. WILLIAMS† GENE H. HOFFNAGLE† † nonvoting member of the Board of STEPHEN L. DIAMOND* Governors BOARD OF GOVERNORS COMPUTER SOCIETY OFFICES IEEE OFFICERS Headquarters Office President: Term Expiring 2004: Jean M. Bacon, Ricardo ARTHUR W. WINSTON Baeza-Yates, Deborah M. Cooper, George V. 1730 Massachusetts Ave. NW Washington, DC 20036-1992 President-Elect: Cybenko, Haruhisha Ichikawa, Thomas W. W. CLEON ANDERSON Phone: +1 202 371 0101 • Fax: +1 202 728 9614 Williams, Yervant Zorian Past President: E-mail: [email protected] Term Expiring 2005: Oscar N. Garcia, Mark A. MICHAEL S. ADLER Grant, Michel Israel, Stephen B. Seidman, Kathleen Executive Director: Publications Office M. Swigger, Makoto Takizawa, Michael R. Williams DANIEL J. SENESE 10662 Los Vaqueros Cir., PO Box 3014 Secretary: Term Expiring 2006: Mark Christensen, Alan Los Alamitos, CA 90720-1314 MOHAMED EL-HAWARY Clements, Annie Combelles, Ann Gates, Susan Phone:+1 714 8218380 Treasurer: Mengel, James W. Moore, Bill Schilit E-mail: [email protected] PEDRO A. RAY Next Board Meeting: 28 Feb. 2004, Savannah, Ga. Membership and Publication Orders: VP, Educational Activities: Phone: +1 800 272 6657 Fax: +1 714 821 4641 JAMES M. TIEN EXECUTIVE STAFF E-mail: [email protected] VP, Publication Services and Products: MICHAEL R. LIGHTNER Executive Director: DAVID W. HENNAGE Assoc. Executive Director: Asia/Pacific Office VP, Regional Activities: MARC T. APTER ANNE MARIE KELLY Watanabe Building Publisher: ANGELA BURGESS 1-4-2 Minami-Aoyama,Minato-ku, VP, Standards Association: JAMES T. CARLO Assistant Publisher: DICK PRICE Tokyo107-0062, Japan VP, Technical Activities: Director, Finance & Administration: Phone: +81 3 3408 3118 • Fax: +81 3 3408 3553 VIOLET S. DOAN RALPH W. WYNDRUM JR. E-mail: [email protected] Director, Information Technology & Services: IEEE Division V Director: ROBERT CARE GENE H. HOFFNAGLE Manager, Research & Planning: JOHN C. KEATON IEEE Division VIII Director: JAMES D. ISAAK President, IEEE-USA: JOHN W. STEADMAN PERSPECTIVES A Tale of Three Disciplines ... and a
Jesse H. Poore University of Tennessee Revolution
espite years of hard work and dedicated champions, software Some disciplines, like engineering still struggles to become the discipline its founders envisioned. Its educational and industrial constituents have circuit and genetic not only failed to converge, but the already large gap between software education and industry practice continues to grow. engineering, seem to CurriculaD are diffuse, software systems continue to have ever larger error counts, and the financial risks of taking on a software project are growing. evolve from theory to Not that efforts haven’t been valiant. Vic Basili, Fred Brooks, Barry practice relatively Boehm, Harlan Mills, and David Parnas, in particular, have worked hard to align software engineering foundations with curricula and curricula with responsibly. Software industry practice. As accomplished scientists and engineers, they have artic- ulated fundamentals, written textbooks, taught in universities, led significant engineering, on the programs in industry, and modeled good practices from positions of pres- tige. Their efforts produced one success after another as software engineer- other hand, while not ing emerged into an academic discipline and profession. yet at the guillotine, has But I believe even they would agree that academic software engineering has had too little impact on the software industry. In part, the evolutionary suffered a decided lack environment was to blame. Like Dickensian protagonists, software engi- neering suffered through the best and worst of times. The best was when its of direction. It may be advocates embraced economic opportunity; the worst was when they strug- time to storm the gates. gled with complexity. Wisdom and foolishness alternated as we pitted for- mal abstractions against individual ingenuity. As the bonds of theory, practice, and tools fused and broke apart by turns, we progressed in the light or wandered aimlessly in the dark. And like the heroes of the French Revolution, we look to a future that will bring us everything or nothing, depending on the public trust. But other disciplines evolved in similar environments and have either become successful or have established the framework to become successful.
THE CHARACTERS Circuit engineering is a discipline that has done the job right, and genetic engineering is at least trying to do the job right. Both have managed vast complexity and achieved a high level of public trust. Software engineering could take lessons from either discipline—and this is not simply an acade- mician’s lament; the cost of software is enormous and the risk to public safety daunting.
30 Computer Published by the IEEE Computer Society 0018-9162/04/$20.00 © 2004 IEEE Circuit engineering Enormous state space characterized this Electronic circuit engineering began with Claude complexity, yet engineers treated it explicitly. From the Shannon’s 1938 thesis. From this point, engineers The number of active components also char- recognized that a circuit was a rule for a function— acterized complexity, which innovators man- beginning, a Boolean function. Pioneers of the field were aged by continually adhering to a description the correspondence strongly motivated to proceed in a thoughtful, of components assembled according to the between circuit methodical way. Electronic components and hard- mathematics of functional composition. diagrams and ware construction were expensive. Repair—destruc- Computer-aided design, engineering, and functional tion and reconstruction—cost both money and time. manufacturing tools were at the industry’s Shaped by talented mathematicians and engi- heart, making abstraction practical and equations neers, the field fully exploited the Boolean algebra enforcing an affinity of theory, practice, and was exact. associated with circuits. In fact, the entire process tools. Testing was an integral part of the engi- of circuit design was a mathematical activity, and neering, and design for testability was a fun- the hardware mirrored the algebraic operations damental consideration. Design tools and AND, OR, and NOT. Theorems made it possible standard best practice were always available and to build any circuit from just NAND gates, which automated in step with hardware evolution. proved economical. Mathematicians developed nor- Reverse engineering was also a consideration, mal forms for functions using specific operators. with its focus on determining a circuit’s function Innovators recognized the power of visual graph- by observing its input-output behavior or even ics early on and devised a notation that the com- divining its internal state design. However, circuit munity adopted and sustained. From the beginning, engineering was and still is mostly a matter of for- the correspondence between circuit diagrams and ward design using established components and functional equations was exact. Only rank ama- leaving a documented design trail. teurs in their home radio shacks would construct As the field progressed, its proponents kept the devices without first sketching diagrams, scribbling research frontier linked to undergraduate training, functions, and consulting parts catalogs. The idea textbook notation, and industry practice. Tools of building every new function from scratch—from were generally in step with demand. The vocabu- AND, OR, and NOT gates—was ludicrous. lary retained the mathematics, physics, and engi- Engineers defined larger units (registers, shift reg- neering concepts that engineers knew and used. isters, adders, multiplexers, demultiplexers, and so There was a consensus as to what constituted good on) mathematically, constructed them physically, professional practice, and practitioners generally and reused them routinely. Reuse was always a managed projects well because everyone under- smart idea. In fact, developers precisely prescribed stood a worker’s qualifications in terms of training and cataloged each component’s functional input- and experience. output relationship for just that purpose. Today, the complexity of circuit engineering This process streamlined the innovation of bet- matches or exceeds that of any other field. ter, cheaper, faster components, since with the func- tional precision of each catalog description, Genetic engineering manufacturers could hide internal secrets from the Although its evolutionary pace is swifter than user. Moreover, the assembly of a new circuit from circuit engineering’s, genetic engineering is follow- components was always clear because a function ing a similar path. The discovery of DNA’s double- represented each component, and all the functions helix structure, attributed to James Watson and were of the same family. The engineer would sim- Francis Crick in 1953, revolutionized life sciences ply formulate the new function at the highest level in general. Although the details vary among organ- of abstraction. isms, we now know that genomes carry a complete Innovations from physics and engineering paced set of instructions for an individual’s lifetime. progress, spanning devices from vacuum tubes to Humans have trillions of cells, each of which has transistors to integrated circuits. At the same time, 23 pairs of chromosomes. Each chromosome has a the mathematical theory flourished. In the 1960s, DNA strand, a regular double-stranded helix as progress accelerated, the theory of finite state linked by bonds between guanine and cytosine on machines as sequence recognizers paved the way the one hand and thymine and adenine on the for scaling up complexity. By the 1980s, circuit other. These linkages are base pairs. The gene itself complexity (still rules for functions) had vastly is a segment of a DNA molecule and can have a increased. few hundred to several million base pairs. The
January 2004 31 human genome, for example, comprises the size of the problem and concluded that their some three billion base pairs. Genetic instruc- only hope was to follow scientific fundamentals Although genetic tions are encoded in these pairs, and like the and best engineering practice, which requires engi- engineering is circuit, provide the rule for a function. neers to catalog and reuse all results and conduct mostly a matter of In genetic engineering, the circuit is already peer reviews of all working methods and work reverse engineering, defined, the program already written; the life- products. In other words, they recognized both a it holds many ware function (or relation) is already imple- sequence-processing problem and a substantial mented in the gene. It is true that some information systems problem. parallels with genetic engineering occurred before the DNA Next, the scientific-industrial leaders of the field circuit engineering. discovery, but the process was trial and error, turned to mathematics to characterize problems guided by experience, intuition, and experi- and to computer science for the appropriate tech- mental principles. The mathematical model nology. Obviously, considering every sequence was for DNA waited just under the surface to be not possible, so scientists adopted a mathematical recognized, and once it was, scientists and engi- approach to successively partition the space and neers embraced it. work productively. The mathematical work is not At this stage, genetic engineering is mostly a mat- complete, but scientists continue to study the alge- ter of reverse engineering. Still, it holds many par- bras and operators. An important remaining task is allels with circuit engineering. Genetic engineers to build sequence recognizers that can divine inner want to determine what a gene does, where it is structure strictly from input-output behavior. located, or which gene causes an observed phe- Complex as these issues are, they are somewhat nomenon (disease or talent). They want to know similar to problems in electrical engineering and how to modify genetic circuits so that the organ- automata theory. In the final analysis, a genetic- ism will respond differently. As with electronic cir- engineering effect is a (possibly nondeterministic) cuits, these modifications can be physically small transformation, and a gene is a rule for a function but produce dramatic behavioral or physical (or relation). Moreover, because most genetic-engi- changes in the organism. neering transformations are embarrassingly paral- Genetic engineers are also interested in learning lel, engineers can use algorithms that exploit how to make new genes from other genes—in parallelism. And finally, researchers who under- essence, how to reuse gene segments as components. stand both mathematics and genetics are writing Lifeware differs from hardware in that it self- the code. Verification and test will proceed care- replicates, can self-repair, uses novel information- fully because errors could lead to false research con- processing modes, and can develop diversity from clusions, harm people and the environment, and undifferentiated state. culminate in a huge financial loss. For the same rea- The life sciences have certainly benefited from sons, the community is highly likely to share mathematical thinking in the past, as in the contri- accepted codes and catalog established results for butions of R.A. Fisher. However, most life scien- subsequent use. tists were not trained in the mathematical The genetic-engineering industry and academia specialties that current field research and practice are committed to the idea that students will be require. Modern genetic engineers must confront trained in the appropriate life science, mathematics, numbers arising from combinatorial complexity on and computer science fundamentals. Curricula a par with the state space that circuit designers rou- changes update the mathematics and computer sci- tinely address. For example, consider all possible ence courses. The industry and research community combinations of base pairs in double strands of evolves tools that enforce genetic engineering’s foun- length 120,000. Then, consider all possible cuts of dations and consensus on processes. The vocabulary length, 1, 2, 3, ..., n, depending on the application retains reference to the mathematics and cell science and all possible reconnections of cuts of compati- on which genetic engineering is built. Industry and ble length that might lead to a crossover having cer- the research community intend to go forward with tain properties. Imagine the complexity if genetic qualified workers trained in the fundamentals (Elias engineers achieve their goals and determine all the Zerhouni, “The NIH Roadmap,” Science, vol. 302, genes in the DNA sequences of multiple organisms 2003, p. 63). and develop intellectual and laboratory tools to use this information with precision. Software engineering The reaction by scientists in genetic engineering Many of us regard the 1968 NATO conference to this situation is exemplary. First, they observed as the birth of software engineering, which was
32 Computer then regarded as the design of computer programs We must work harder at becoming disci- and software-intensive systems within a perfor- plined: Whereas physics and hardware reali- We can mance, quality, and economical framework. ties hold circuit engineering to fundamentals, design software (Ironically, the source material for Frederick P. and chemistry and life imperatives restrict Brooks’ The Mythical Man Month was shipping in genetic engineering, software engineering for testability— quantity at that time, so perhaps the conference must freely elect and self-impose its intellec- but we assumed that it is never too late for a revolution.) tual principles. Nonetheless, the complexities generally don’t. Harlan Mills was the keynote speaker at the 1975 are more or less equal. ACM national conference, where he presented “The Complexity in the software industry grew New Math of Computer Programming” (Comm. more slowly than in genetic engineering, yet ACM, Jan. 1975, pp. 43-48)—one of his many today, the genetic engineering industry is ahead efforts to remind us of the mathematical basis for of software engineering in principles, methods, software development. Like a circuit, a program is training, notation, vocabulary, tools, and produc- a rule for a function and subject to an algebra of tivity! Why? Because both genetic engineering and program construction, which Mills defined. All the circuit design follow good science and engineering benefits that accrued to the circuit design industry practices. could have accrued to the software industry had Software engineering had a later start than either researchers and practitioners simply acted similarly circuit or genetic engineering, so it had the chance on this basic fact. to model itself after either one. We knew from John Yet, despite our knowledge, the software indus- von Neumann that software development is an try progressed by dint of nonconformist intellect inherently mathematical activity—just as much so and massive amounts of money rather than by dis- as circuit or genetic engineering—and we under- ciplined science and engineering. stood many of the mathematical facts from the out- set. We grasped that we ask the same questions as A FAR BETTER THING circuit or genetic engineers: What must the new In each of the three disciplines, the steps from program do? What does this program do? Where conception to final product are similar. For specifi- is the program that does such and such? How can cations in circuit engineering, we have functions we modify the program to behave differently or over fixed-length sequences of a two-symbol alpha- construct new programs more efficiently? bet with Boolean axioms. For genetic engineering, Ironically, software engineering uses little com- we have functions over arbitrary-length sequences puter power relative to circuit engineering or of a four-symbol alphabet with DNA axioms. For genetic engineering to support design and testing. software engineering, we have functions over arbi- All three disciplines are (or could be) based on very trary-length sequences of an arbitrary alphabet similar symbol manipulation, component reposi- with axioms that must be derived from the appli- tories, design and validation, state space manage- cation itself. ment, and certification testing. In all three cases, specifications give rise to state Circuit and gene designers are generally willing to machines. Prescriptions for implementation in var- pay the computational price for design and testing, ious systems, or the code, map the state machine to but software engineering will cop a plea rather than the target implementation. Code can be correct by turn to supercomputers when the state space grows design in all cases. We can design circuits with testa- large or (statistically) exhaustive testing is in order. bility built in. We can design genes with a known This may sound harsh, but we have much at stake. way to test them. And we can design software for I’m not denying the software accomplishments, testability—but we generally don’t. When we can’t the existence of virtuoso programmers and effec- fully automate exhaustive testing, we can use sta- tive development teams, or the use of tools such as tistical models to assure that the circuit/gene/soft- editors, compilers, and leak detectors. (Indeed, ware meets quality standards by statistical protocol. many have debunked the “productivity paradox.”) Like circuit and genetic engineers, software engi- But neither am I willing to ignore the chaos, fail- neers should look first to the underlying science for ures, and risks to public safety nor the enormous vocabulary, methods, and tools. Half a century sums wasted. later and after a correct start, the software indus- We can manage only so much complexity ad hoc, try remains disconnected from university training, and we can absorb only so much financial waste and university training is disconnected from soft- and tolerate only so much risk to public safety. We ware’s mathematical foundations. have compelling reasons to try to get it right, espe-
January 2004 33 cially where correctness, performance, qual- mia’s early efforts to lead gave way to frantic efforts ity, and economy really matter. to appear relevant to an industry unencumbered by Erroneous and The software engineering industry would the baccalaureate experience. arbitrary solutions be better off today if it had followed the pat- This failure has many ugly faces, and the list of are too often tern of circuit engineering, and it will be bet- have-nots is long: We have no accepted as the ter off tomorrow if it follows the path of genetic engineering. I find it bizarre that, • standard notation with mathematical fidelity designer’s creative given software’s origin and its very nature, that everyone will use, prerogative. we could have such a gross disconnection • normal forms and cataloged components, among the field’s foundations, university cur- • experimentally validated best practices ricula, and general practice. We can only par- • productive programming languages in wide- tially blame the lack of physical or chemical spread use, constraints, legions of extremely bright indepen- • tools evolved systematically over decades that dent thinkers, and unlimited sums pushed into soft- facilitate abstraction and enforce software’s ware development and maintenance. Also, the foundations in industrial practice, industry has become complacent, enjoying an inex- • vocabulary with a mathematical and engi- plicable degree of blind public trust and acceptance neering basis (instead we have beans), of low quality. It has had little motivation to con- • curricula in close rapport with the workforce, nect practice with preparation. • standards for product usability and correct- ness, MADAME DEFARGE’S SCARF • workforce trained in the fundamentals and Every revolution starts with a list of culprits that experienced in standard practice based on fun- must be beheaded or at least retrained. In software damentals, and engineering, the culprit is our academic programs. • industry that will employ only qualified work- Across the university, from accounting to zoology, ers and pay only for quality work. we are teaching students to be proud of writing small programs badly (comparatively few of the Software is the hope and means for progress in population design little circuits and little genes) and science and in better industrial products, yet soft- to be grateful for shabby application software. ware engineering seems to be languishing in a dark I recently examined several of the software pack- prison of its own making. ages included with nutrition textbooks as well as software that health clinics actually use. I found VIVE LA RÉVOLUTION that although the textbooks had a rather uniform It will take much hard work and planning to get look and feel, the software packages were very dif- academia and industry into alignment and make ferent, ranging from intuitive and easy to use to software engineering worthy of public trust. I don’t inscrutable. The programs were easy to hang and see that we have a choice. We can’t afford not to most lost state awareness during use. Moreover, try. Worldwide software industry waste is measured when I entered the same data using each package, in hundreds of billions of dollars annually, exclud- I got materially different results. We seem to have ing physical (death) and social (identity theft) costs. no quality watchdog for this kind of software, since Must heads roll before we do something? book production attends to the book’s quality but To start this revolution, we must advance three ignores the software that comes with it. fronts simultaneously. In our primary computer science, computer engi- neering, information science, information engi- Certification neering, and even software engineering curricula, Professional societies, industry associations, and we fail to introduce and reinforce with urgency the government regulators must certify products in nar- mathematical foundations throughout the degree row specializations. We could begin with our old- program: disciplined system and program design, est specialty, mathematical software. There is a the seriousness of design errors, and designs that worldwide trust of numerical analysts who have a satisfy a social responsibility, not just the author’s deep understanding of algorithms, performance ego. Erroneous and arbitrary solutions are too often and error in calculating transcendental functions, accepted as the designer’s creative prerogative. numerical integration and differentiation, basic lin- A look at history from the 1965 ACM curricu- ear algebra operations, and so on. People in the lum to current offerings makes it clear that acade- know can name the mathematicians, software,
34 Computer hardware, and organizations to be trusted. It would certifying that the software’s authors are be a relatively simple step for this society to define dependable, capable, and working according scientific certification protocols for such software- to an approved process for the application in Programming hardware implementations and to issue a seal of question. Airline pilots and mechanics require language compilers approval for each use by a compiler or application formal training and a license, but not so for is an area that is package. We could then relax and trust the calcu- the authors of the avionics software systems sufficiently mature lations behind a spreadsheet GUI or application that fly the planes. Pharmaceuticals must stand to support effective package. This is especially important when the up to statistical protocols in clinical trials, industry experiences major shifts, for example, but the software systems that aid in the man- certification when the mainframes of the 1970s gave way to the ufacture, distribution, and tracking of these protocols. workstations of the 1980s and when vector and pharmaceuticals go unchecked. parallel processors arrived on the scene. The examples are endless: digital controls Programming language compilers is another area of nuclear power plants, air-traffic-control that is sufficiently mature to support effective cer- systems, automotive braking and (soon-to-come) tification protocols. Existing professional societies steering systems, and so on and so on. could step up to this task. In view of the 50 years of programming language research and the poten- Curriculum reform tial for accumulating knowledge and skill in com- While certification and licensing apply pressure piler implementation, we have little justification for from the top down, curriculum reform attacks using languages that fail to enforce sound practices the problem from the bottom up. McMaster Uni- or for tolerating compiler errors or variations versity’s software engineering curriculum is the best among compilers of the same language. example of how to align foundations with curric- Industry associations could establish certification ula and curricula with engineering practice (David protocols and centers for many consumer products L. Parnas, “Software Engineering Programmes Are that are software intensive but not a threat to pub- Not Computer Science Programmes,” Ann. Soft- lic safety. Perhaps the controls for home appliances ware Eng., vol. 6, 1998, pp. 19-37). The curricu- would become simpler, more uniform, and less lum holds that software developers should prone to transcendental behavior. understand the products or systems they are devel- In one certification area—medical devices and oping. It also relies on an understanding of physi- other products that directly relate to public safety— cal systems and control engineering to exert we should begin immediately. Government regula- discipline on the software. Finally, it focuses on tory agencies may not do this type of certification knowing and teaching what is true of software. expertly at first, but in time they should be able to I sincerely hope that after apprenticeships under perform it at least as well as they now do for pre- masterful software engineers, graduates of the scription drugs and Florida orange juice. McMaster program will be armed with an advan- A few other areas—statistical packages built on tage beyond knowledge, more than a college degree mathematical software, transaction processing and and more than successful experience—an engi- database management systems, and embedded real- neering license that permits the holder to do things time systems—are also mature enough for a mean- that those without the license cannot do. ingful measure of product certification by cohorts The process of declaring individuals qualified or of well-qualified practitioners. unqualified to participate in the development of specific software-intensive products would go a Licensing long way toward driving consensus among acade- Regulatory authorities must license certain prac- mics worldwide to get the curricula right. titioners and processes, such as those associated with Curricula reflect the degree of disciplinary con- medical devices and financial systems. Software- sensus and some degree of local circumstances and intensive systems now dominate many devices and strategy. If there are three schools of thought, there processes that impinge on social and physical pub- will be three flavors of curricula. Accreditation and lic well-being. The risks are unacceptably high. standards are both a blessing and a curse: If applied Surgeons require formal training and a license, but too soon, they stifle development and creativity; if the authors of the software inside surgical instru- applied too late, the field becomes woefully sub- ments do not. Software runs medical devices from optimal. If applied first to the best programs, they pacemakers to radiation-therapy systems, but no become goals for all to achieve; if applied first to public authority is thoughtfully and conservatively weak programs, they become company to avoid.
January 2004 35 Curriculum reform is, of course, a matter the same time differentiate some software devel- of building consensus where we have too lit- opment methods as superior in quality and pro- The software tle. Mathematics is taught across the univer- ductivity to the efforts of those unschooled in the field must find sity, often to avoid foundations and treat foundations. We must establish both in fact and a way to bind applications directly, but a hard problem is message, “Our way isn’t just different; it’s better in university gladly taken to a member of the mathemat- ways that matter.” ics faculty for expertise and assistance. It is fair to ask whether circuit engineering and preparation with English is used and taught across the univer- genetic engineering resorted to certification and industry practice. sity, but a member of the English faculty will licensing. In some sense the answer is yes. There are be the final arbiter of a dispute on grammar, professional engineering licenses, home wiring however pedantic the experience might be. codes, appliance design standards, and indepen- Electrical engineering is the place to get dent laboratory testing. When electrical shock is advice on circuit design, and the life sciences own not a threat, the threat of product recall drives high genetic engineering. Computer and information sci- standards or, more recently, drives the functional- ence and engineering, on the other hand, seem to ity from hardware into software. Because of soci- be in everyone’s job description. Worse, when a etal concern for genetically altered foods, the role good physicist becomes a bad software developer, of genetics in medical therapies, and the public fear it is with unstinting confidence and resolve that he of uncontrolled experimentation with life forms, knows no peer. Thus, it is no surprise that when the we may see extensive government involvement in nutritionist encounters an array of software pack- most aspects of genetic engineering—perhaps all ages that supposedly do the same job but seem to except the software. have an arbitrary design, he wonders what the com- We have seen with circuit engineering—and will puter scientists are teaching. see with genetic engineering—close attention to Curriculum reform must start with the freshman preparing students for field participation and care- service courses. We must communicate the essence ful integration of workers in industrial practice. and importance of the field’s foundations and at The software field must find a similar way to bind university preparation with industry practice. New! t is never too late for a revolution, and in cir- Software Engineering Overview: cuit and genetic engineering we have two wor- I thy role models. Given the state of affairs in Preparing for the software engineering today and the pervasive and IEEE Computer Society intrusive role of high-risk software in public wel- fare, I, for one, am willing to be a Sydney Carton— CSDP Examination to act quickly to save something I cherish. In this case, the price is one of calculated restriction to save Expand your knowledge and advance your the reputation of software engineering and ensure that it is never the weakest link. Theoretically, soft- career with a new online training course ware is the only component that can be perfect, and created exclusively for software professionals. this should always be our starting point.
This interactive course offers 10 modules that cover all Jesse H. Poore is a professor of computer science at the University of Tennessee and holds the 11 knowledge areas of the CSDP exam. Key concepts and Ericsson-Harlan D. Mills Chair in Software Engi- principles of software engineering are presented with neering. He is the Director of the University of Ten- clear and concise explanations. Plus the course provides nessee–Oak Ridge National Laboratory Science a list of references to aid further study for each area. Alliance. His current interests are the economical production of high-quality software and federal sci- ence policy. Poore received a PhD in information Register Now and computer science from the Georgia Institute of For more information or to enroll for Technology. He is a member of the IEEE, the IEEE Software Engineering Overview online by visit Computer Society, and the ACM and is a fellow of the AAAS. Contact him at [email protected]. www.computer.org/certification/distancelearning PERSPECTIVES
The End of Science
John Horgan Revisited
ne of my most memorable moments as a science journalist Because science has occurred in December 1996, when I attended the Nobel Prize festivities in Stockholm. During the white-tie banquet for advanced so rapidly 1,500 presided over by Sweden’s King and Queen, several prizewinners stood to give brief speeches. David Lee of Cor- over the past century nellO University, a winner in physics, decried the “doomsayers” who were claiming that science is ending; the work for which he and his colleagues or so, we assume that it had been honored showed just how vital physics is. can and will continue As the audience applauded, several people at my table whispered and jabbed their fingers at me. They knew Lee was alluding to my book, The to do so, possibly End of Science (Addison-Wesley, 1996), which had been stirring up trouble since its release six months earlier. The book argued that science—especially forever. But science pure science, the grand quest to understand the universe and our place in it— might be entering an era of diminishing returns. As the New York Times itself tells us that there put it in a front-page review, “The great days of scientific discovery are over; are limits to our what science now knows is about all it will ever know.”1 The book was denounced by eminences such as Bill Clinton’s science advi- knowledge. sor, the British minister of science, the heads of NASA and the Human Genome Project, the editors of Science and Nature, and dozens of Nobel laureates. These denunciations usually took the form not of detailed rebut- tals of my arguments but of declarations of faith in scientific progress. Scientists need a certain degree of faith to bolster their confidence in the arduous quest for truth; lacking such faith, science would not have come so far so fast. But when researchers reflexively deny any evidence and argu- ments that challenge their faith, they violate the scientific spirit. Perhaps recognizing this fact, some pundits reacted thoughtfully to my book. Moreover, I suspect that more than a few scientists who publicly assailed my views privately acknowledged their merit. David Lee was a case in point. When I introduced myself to him at the Nobel banquet to tell him how flattered I was that he had mentioned my book, he said he hoped that I hadn’t been offended. He had enjoyed the book and had agreed with much of it, particularly the argument that achieving fundamental discoveries is becoming increasingly difficult. He just felt that the only way for scientists to truly know the limits of science is to keep trying to overcome them.
BUMPING UP AGAINST LIMITS Today, scientists and nonscientists alike still find it hard to accept that sci- ence may be bumping up against limits. I can understand why. We have grown
0018-9162/04/$20.00 © 2004 IEEE Published by the IEEE Computer Society January 2004 37 up in a period of explosive scientific and tech- has an intriguing corollary: If science has limits, The more we know nological progress, reflected by such measures then it might be moving at maximum speed just as Moore’s law. When I started working as a before it hits the wall. about the Earth, journalist in 1983, I wrote on an IBM the less there is Selectric typewriter. I literally cut and pasted NANOTECH AND FUSION to discover. (actually Scotch-taped) manuscripts, and Some researchers grant that the basic rules gov- slathered Whiteout on like housepaint. My erning the physical and biological realms may be first computer, a “portable” Kaypro, was as finite, and that we may already have them more or bulky as a filing cabinet and had a 64K mem- less in hand. But they insist that we can still explore ory. Ten years later, I was writing The End of Science the consequences of these rules forever and manip- on a Macintosh laptop. Talk about progress! ulate them to create an endless supply of new mate- Because science has advanced so rapidly over the rials, organisms, technologies. Proponents of this past century or so, we assume that it can and will position often compare science to chess. The rules continue to do so, possibly forever. But science itself of chess are quite simple, but the number of possi- tells us that there are limits to our knowledge. ble games that these rules can give rise to is virtu- Relativity theory prohibits travel or communica- ally infinite. tion faster than light. Quantum mechanics and This point is reasonable, but some enthusiasts— chaos theory constrain our predictive ability. particularly in the trendy field of nanotechnology— Evolutionary biology keeps reminding us that we take it too far. As espoused by evangelists such as are animals, designed by natural selection not for Eric Drexler, nanotechnology resembles a religion discovering deep truths of nature but for breeding. more than a field of science.3 Drexler and others Perhaps the most important barrier to future proclaim that we will soon be able to reconstruct progress in science—especially pure science—is its reality from the atomic scale on up in ways limited past success. only by our imaginations. Our alchemical power Postmodernist philosophers will find this a terri- to transform matter will help us achieve infinite bly naive comparison, but scientific discovery in wealth and immortality, among other perks. some respects resembles the discovery of the Earth. Nanotechnology has also inspired some enter- The more we know about the Earth, the less there is taining science fiction about nanobots running to discover. We have mapped out all the continents, amok, such as Michael Crichton’s novel, Prey oceans, mountain ranges, and rivers. Every now and (Harper Collins, 2002), and an essay by Bill Joy of then, something interesting turns up. Scientists find Sun Microsystems titled “Why the Future Doesn’t a new species of lemur in Madagascar or exotic bac- Need Us” (Wired, Apr. 2000). But to the extent that teria living in deep-sea vents. But at this point we are researchers have experimental experience at the unlikely to discover something truly astonishing, like nanoscale level, they tend to doubt the more far- the lost continent of Atlantis or dinosaurs dwelling fetched claims—positive or negative—that nan- inside the Earth. otechnologists make. “The level of hard science in In the same way, scientists might be unlikely to these ideas is really low,” the Harvard chemist discover anything surpassing the big bang, or George Whitesides remarked recently.4 After all, quantum mechanics, or relativity, or natural selec- nanotechnology is just a glossy wrapping for nitty- tion, or DNA-based genetics. Nobel Prizes reflect gritty work done in chemistry, molecular biology, the trend toward diminishing returns. The Russian solid-state physics, and other fields that investigate physicist Pyotr Kapitsa discovered superfluidity nature at small scales. Experiments often reveal that in liquid helium in 1938 and won a Nobel Prize what should work in principle fails in practice. for that finding 40 years later. David Lee and his Take nuclear fusion. Physicists such as Hans two colleagues won the 1996 prize for showing Bethe elucidated the basic rules governing fusion— that superfluidity also occurs in a helium isotope, the process that makes the sun and other stars He-3. shine—more than 60 years ago. By the 1950s, this If we accept that science has limits—and science knowledge had spawned the most fearsome tech- tells us that it does—then the only question is when, nology ever invented: thermonuclear weapons. not if, science reaches them. The American histo- Next, physicists hoped to harness fusion for a more rian Henry Adams observed a century ago that sci- benign application: a clean, economical, boundless ence accelerates through a positive feedback effect.2 source of energy. Knowledge begets more knowledge; power begets When my career began in the early 1980s, more power. This so-called acceleration principle fusion-energy research was a staple of the science
38 Computer beat: Keep the money coming, researchers sory randomness “pseudonoise.” Two par- promised, and in 20 years we will give you energy adigmatic examples of pseudonoisy systems AI researchers have too cheap to meter. Twenty years later, the US are the Mandelbrot set, discovered by Benoit has drastically reduced its budget for magnetic- Mandelbrot, and “Life,” a cellular automa- come up with some confinement fusion, formerly the leading candidate ton devised by John Conway. useful inventions, for power generation. Chaoplexologists such as Wolfram assume but these advances Inertial-confinement fusion—in which giant lasers that much of the noise that seems to pervade pale beside the blast tiny fuel pellets—has fared better, but only nature is actually pseudonoise, the result of hopes that because its main application is nuclear-weapons some underlying, deterministic algorithm. research. Surveys of sustainable-energy methods This hope has been nurtured by certain gen- AI pioneers once rarely even give fusion a courtesy mention any uinely significant findings, such as Mitchell had for their field. more—and please don’t bring up cold fusion. Feigenbaum’s discovery more than 20 years Diehards may still cling to the dream, but realists ago that gushing faucets and similar turbulent acknowledge that fusion energy is effectively dead, systems, although they seem hopelessly noisy, the victim of technical, economic, and political con- actually adhere to a rule called period doubling. straints. But a gushing faucet is laughably simple com- pared to a stock market, a human brain, a genome, COMPUTERS AND CHAOPLEXITY or a rain forest. These hideously complex phe- Of course, technological advances often enable nomena—with their multitudes of variables—have researchers to overcome seemingly insurmountable shown no signs of yielding to the efforts of chao- obstacles. Computers in particular have vastly plexologists. One reason may be the notorious but- increased scientists’ capacity for data acquisition, terfly effect, elucidated by Edward Lorenz in the analysis, storage, and communication. Innovations 1960s. The butterfly effect limits both prediction such as optical and quantum computing may and retrodiction, and hence explanation; specific extend the reign of Moore’s law indefinitely— events cannot be ascribed to specific causes with although, as the astute computer theorist Rolf complete certainty. This is something that has Landauer often warned, quantum computers may always puzzled me about chaoplexologists: be so sensitive to thermal noise and other disrup- According to the butterfly effect—one of their fun- tions that they remain a laboratory curiosity.5 damental tenets—achieving many of their goals But adherents of certain computer-driven fields— may be impossible. notably artificial life, chaos, and complexity—seem to view computers not as tools but as wands that SEEKING ARTIFICIAL COMMON SENSE will magically solve even the toughest puzzles. In One would think that chaoplexologists would also The End of Science I lumped chaos and complex- be chastened by the failure of artificial intelligence ity together under a single term, chaoplexity, to live up to expectations. AI researchers have come because, after talking to scores of people in both up with some useful inventions, including devices fields, I realized there is no significant difference that can translate languages, recognize voices, judge between them. loan applications, interpret cardiograms, and play Chaoplexologists have argued that with more chess. But these advances pale beside the hopes powerful computers and mathematics they can that AI pioneers once had for their field. solve conundrums resistant to conventional scien- In 1984, I edited an article for IEEE Spectrum in tific reductionism, particularly in “soft” fields such which AI expert Frederick Hayes-Roth predicted as ecology, psychology, economics, and other social that expert systems were going to “usurp human sciences. Stephen Wolfram recently reiterated these roles” in professions such as medicine, science, and claims in his magnum opus, A New Kind of the business world.7 When I called Hayes-Roth Science, which touts cellular automatons as the key recently to ask how he thought his predictions had that will unlock all the riddles of nature.6 held up, he cheerfully admitted that the field of As many critics have noted, Wolfram’s “new” expert systems, and AI generally, had stalled since science actually dates back at least to John von their heyday in the early 1980s. Neumann, the inventor of cellular automatons. Not all AI’ers have conceded defeat. Hans Von Neumann and many others have shown that Moravec and Ray Kurzweil still prophesize that simple rules, when followed by a computer, can machines will soon leave flesh-and-blood humans generate patterns that appear to vary randomly in their cognitive dust.8,9 (In The End of Science, I as a function of time or scale. Let’s call this illu- called this sort of speculation “scientific theology.”)
January 2004 39 These cyberprophets cite Garry Kasparov’s viewed him in 1997, Brooks said, “Ultimately you There are clear loss to the IBM computer Deep Blue in 1997 have to ground it out. You have to attach it to some signs that as a portent of AI’s impending triumph. other sensory motor experience, and that’s what I Actually, that contest underscored the lim- think he’s missing.” mathematics is itations of artificial intelligence. Chess, with already outrunning its straightforward rules and tiny, Cartesian THE END OF MATHEMATICS? our limited cognitive playing field, is a game tailor-made for com- Some of the harshest attacks on The End of capabilities. puters. Deep Blue, whose five human handlers Science came from scientists whose fields I hadn’t included the best chess programmers in the bothered to denigrate. world, was a prodigiously powerful machine, When I encountered him at a New York Academy capable of examining hundreds of millions of of Sciences meeting in November 1996, chemist and positions each second. If this silicon monster had to Nobel laureate Dudley Herschbach commented that strain so mightily to beat a mere human at chess, not only had I failed to include a chapter on chem- what hope is there that AI engineers will ever create istry, my index included only two measly references HAL, the lip-reading killer in the film 2001? to the subject. I didn’t have the heart to tell Some prominent AI’ers acknowledge that com- Herschbach that chemistry just seemed too passé to puters will probably never be as smart as HAL. dwell on. In a 1992 interview, Linus Pauling assured Although computers excel at tasks that can be pre- me that he had laid out the basic principles by 1930, cisely defined, such as chess, they will never acquire and who was I to disagree with Linus Pauling? (Also, the flexible, all-purpose intelligence—the ordinary to be honest, I have always found chemistry excru- common sense—that every normal human acquires ciatingly dull.) by the age of five or so. Similarly, the mathematician John Casti com- In HAL’s Legacy: 2001’s Computer as Dream plained in Nature12 that I had neglected to include and Reality,10 a recent collection of essays by AI a chapter titled “The End of Mathematics.” experts, David Kuck stated flatly, “Under any gen- Actually, I had planned to include such a chapter; eral definition ... AI so far has been a failure.” I just ran out of gas. I would have agreed with Casti Roger Shank declared that HAL “is an unrealistic that there are no limits, in principle, to mathemat- conception of an intelligent machine” and “could ics, because mathematics is a process of invention never exist.” The best that computer scientists can rather than of discovery; in that sense, mathemat- hope to do is to create machines “that will know a ics resembles art or music more than pure science. great deal about what they are supposed to know Moreover, Kurt Godel’s incompleteness theorem about and miserably little about anything else.” established that any moderately complex system of Even Marvin Minsky, who had predicted in the axioms gives rise to questions that cannot be mid-1960s that computers would be as smart as answered with those axioms. By adding to their humans within three to eight years, admitted that base of axioms, mathematicians can keep expand- “we really haven’t progressed too far toward a truly ing the realm within which they play, posing new intelligent machine.” conjectures and constructing new proofs, forever. One AI’er still pursuing the original vision of a The question is, how comprehensible will these computer with an all-purpose rather than highly spe- proofs be? cialized intelligence is Douglas Lenat. For 20 years, There are clear signs that mathematics is already he has been trying to create a software program that outrunning our limited cognitive capabilities. The mimics common-sense knowledge. Lenat’s goal was largest conventional proof ever constructed is the for this program, called Cyc, to become more or less classification of finite simple groups, also called autonomous, capable of acquiring new knowledge “the enormous theorem.” In its original form, this by scanning newspapers, books, and other sources theorem consisted of some 500 separate papers, of information. In 1997, Lenat predicted that by totaling more than 20,000 pages, written by more 2001 Cyc would become a “full-fledged creative than 100 mathematicians over a period of 30 years. member of a group that comes up with new discov- It has been said that the only person who really eries. Surprising discoveries. Way out of boxes.”11 understood the proof was Daniel Gorenstein of The world is still waiting. Rodney Brooks of MIT Rutgers University, who served as a kind of gen- has complained that Cyc’s intelligence has nothing eral contractor for the project. Gorenstein died in in common with the human variety; far from inter- 1992. acting with the world in flexible or creative ways, A growing number of mathematical proofs are Cyc is really just a fancy dictionary. When I inter- constructed with the help of computers, which can
40 Computer carry out calculations far beyond the capability of true, empirical science. Ironic science crops mere mortals. The first such proof, constructed in up in the so-called hard sciences, such as 1976, demonstrated the truth of the four-color physics and cosmology. Superstring theory Neuroscience’s most theorem, which states that four hues are sufficient is my favorite example of ironic science. It’s important discovery to color even an infinitely broad map so that no science fiction with equations. may be that different identically colored countries share a border. The But ironic science is most pervasive in fields regions of the brain proof depended on a calculation that took a com- that address human cognition and behavior. are specialized puter 1,000 hours to complete. Theories of human nature never really die— Mathematicians like to wax rhapsodic about the they just go in and out of fashion. Often, old for carrying out elegance, beauty, and depth of proofs. But computer ideas are simply repackaged. The 18th-cen- different functions. proofs yield truth without insight or understanding. tury pseudoscience of phrenology is reincar- So yes, mathematics can, in principle, continue for- nated as cognitive modularism. Eugenics ever. The problem is that no mere human will be evolves into behavioral genetics. Social able to understand it. In 1997, the mathematician Darwinism mutates into sociobiology, which in Ronald Graham said to me,“We’re not very well turn is reissued as evolutionary psychology. adapted for thinking about the space-time contin- One astonishingly persistent theory is psycho- uum or the Riemann hypothesis. We’re designed for analysis, which Freud invented a century ago. Once picking berries or avoiding being eaten.” defined as “the treatment of the id by the odd,” psy- choanalysis has been subjected to vicious criticism THE UNDISCOVERED MIND since its birth. Some Freud-bashers imply that while Some criticism of The End of Science, I admit, was French philosophers and other fuzzy-brained sorts on target. When I met the eminent British biologist may still fall under Freud’s spell, real scientists are Lewis Wolpert at a biology conference in London in immune to his charms. Actually, many prominent 1997, he declared that my book was “appalling, neuroscientists—such as Nobel laureates Gerald absolutely appalling!” He was particularly upset by Edelman13 and Eric Kandel14—still defend Freudian the chapter titled “The End of Neuroscience.” How theory. dare I dismiss all the vast and vital research on the Freud’s ideas have persisted not because they brain in a single chapter, which focused not on gen- have been scientifically confirmed but because a uine neuroscientists but on Francis Crick, a molec- century’s worth of research has not produced a par- ular biologist, and Roger Penrose, a physicist? adigm powerful enough to render psychoanalysis Neuroscience was just beginning, not ending! obsolete once and for all. Freudians cannot point to Wolpert stalked away before I could tell him that unambiguous evidence of their paradigm’s superi- I thought his objection was fair. I had already ority, but neither can proponents of more modern decided that mind-related science had so much paradigms. potential to alter our world, both intellectually and materially, that it deserved a more serious, detailed THE HUMPTY DUMPTY DILEMMA critique than it got in The End of Science. I offered Neuroscience was supposed to deliver us from such a critique in my second book, The Un- this impasse. Neuroscientists have acquired an discovered Mind: How the Brain Defies Repli- astonishing ability to probe the brain with micro- cation, Medication, and Explanation (Free Press, electrodes, magnetic resonance imaging, positron- 1999). As the subtitle suggests, this book looks at emission tomography, and other tools. Neuro- attempts to explain the mind, treat its disorders, science is clearly advancing. It’s getting somewhere. and replicate its functions in computers. In addi- But where? So far, neuroscience has had virtually tion to neuroscience, this book also covers psy- no payoff in terms of diagnosing and treating such chiatry, psychopharmacology, behavioral genetics, complex mental illnesses as schizophrenia and evolutionary psychology, and artificial intelligence. manic depression. It has failed to winnow out all My conclusion was that these fields have largely the competing unified theories of human nature, failed to live up to their advertising. In spite of all whether psychoanalysis, behaviorism, connection- this investigation, the mind remains largely undis- ism, or evolutionary psychology. covered. In The End of Science, I coined the term Neuroscience’s most important discovery may be “ironic science” to describe science that never gets that different regions of the brain are specialized a firm grip on reality and thus doesn’t converge for carrying out different functions. For example, on the truth. Ironic science is more like philoso- the visual cortex contains one set of neurons dedi- phy, literary criticism, or even literature than like cated to orange-red colors, another to objects with
January 2004 41 high-contrast diagonal edges, and still Rational Mysticism (Houghton Mifflin, 2003). Mystical another to objects moving rapidly from left While researching the book, I spoke to a wide vari- experiences to right. The question is, how does the brain ety of scholars studying mysticism, including neu- coordinate and integrate the workings of roscientists, psychologists, and anthropologists. can’t give us the these highly specialized parts to create a Early on, one of these sources warned me that kind of absolute mind? Neuroscientists have no idea. This is you can’t comprehend mystical experiences if knowledge that sometimes called the binding problem, but you’ve never had one. With that in mind, I learned we crave. I prefer to call it the Humpty Dumpty Zen meditation. I had my temporal lobes electro- dilemma. Neuroscientists can take the brain magnetically tickled by a device called the “God apart, but they can’t put it back together machine.” I consumed a nauseating psychedelic again. brew called ayahuasca, which serves as a religious Particle physicists once faced a similar dilemma. sacrament for Indians in the Amazon. I concluded In the 1950s, the number of particles detected in that mystical experiences can’t give us the kind of accelerators proliferated wildly. Theorists trying to absolute knowledge that we crave. Quite the con- make sense of it all were baffled. Then a brilliant trary. Mystical experiences do not give us The young physicist named Murray Gell-Mann showed Answer to the riddle of our existence; rather, they that all these different particles were made of a few let us see just how truly astonishing the riddle is. more fundamental particles called quarks. Order Instead of a big “Aha!” we get a big “Huh?” emerged from chaos. But in terms of sheer complexity, particle physics WHAT’S MY POINT? is a child’s game compared to neuroscience. All pro- After I gave a talk on the limits of science at tons, neutrons, and electrons are identical; a the- Caltech, a neuroscientist in the audience angrily ory that applies to one particle applies to all. But asked me what my point was. Did I think he and his each brain is unique, and it changes every time its colleagues should simply give up their research? owner is spanked, learns the alphabet, reads Thus Should Congress take its funding away? Good Spake Zarathustra, takes LSD, falls in love, gets questions. divorced, undergoes Jungian dream therapy, or suf- I hope I don’t sound disingenuous when I say fers a stroke. Scientists cannot simply ignore each that I would hate to see my prophesies become self- individual’s uniqueness, because it is central to our fulfilling—not that there is any chance of that any- humanity. This fact immensely complicates the way. I always encourage young people to become search for a unified theory of the brain and mind. scientists, and I would be delighted if my children Some scientists have reluctantly concluded that pursued that path someday. science may never fully solve the mysteries of the Grant for a moment that I am right—that sci- brain and mind. This position is sometimes called ence will never again yield revelations as monu- “mysterianism.” One well-known mysterian is the mental as the theory of evolution, general rela- psychologist Howard Gardner, who contends that tivity, quantum mechanics, the big bang theory, neither psychology, neuroscience, nor any other DNA-based genetics. For example, physicists will field has provided much illumination of such peren- never discover a unified theory that reveals “the nial riddles as consciousness and free will.15 These mind of God,” as Stephen Hawking once put it.16 subjects “seem particularly resistant to [scientific] And grant that some far-fetched goals of applied reductionism,” Gardner said. He suggests that psy- science—such as immortality, superluminal space- chologists may advance by adopting a more “liter- ships, and superintelligent machines—may forever ary” style of investigation and discourse—the style elude us. that Freud exemplified. I nonetheless have no doubt that researchers will A surprising number of scientists have proposed find better treatments for cancer, schizophrenia, that investigations of mystical states of conscious- AIDS, and other diseases; more benign sources of ness—such as those induced by meditation, prayer, energy (other than fusion); and more convenient or psychedelic drugs—might yield insights into the contraception methods. In the realm of pure sci- mind that complement or transcend those of sci- ence, scientists will surely gain a better under- ence. This notion has inspired a series of highly standing of how galaxies formed, how life began publicized meetings—most recently at MIT last on Earth, how Homo sapiens became so smart so September—between prominent scientists and the fast, how neural processes generate awareness, how Dalai Lama, the leader of Tibetan Buddhism. I a single fertilized egg turns into a fruit fly or a con- explore this possibility in my most recent book, gressman.
42 Computer But I would like to see a greater recognition of It or Praise It?” New Ideas in Psychology, vol. 10, science’s limitations—particularly in mind-related no. 2, 1992, pp. 179-190. fields, where our desire for self-knowledge can 16. S.W. Hawking, A Brief History of Time, Bantam make us susceptible to pseudoscientific cults such as Books, 1998. Marxism, social Darwinism, eugenics, and psy- choanalysis. Science is never more dangerous than when it seeks to tell us what we are, what we can be, and even what we should be. John Horgan, a former senior writer at Scientific American, has written for the New York Times, Discover, Slate, Time, and other publications. y goal is to foster an attitude that I call He has posted essays, articles, and profiles at “hopeful skepticism.” Too much skepti- www.johnhorgan.org. Contact him at jhorgan@ M cism culminates in a radical postmod- highlands.com. ernism that denies the possibility of achieving any truth. Too little skepticism leaves us prey to ped- dlers of scientific snake oil. But just the right amount of skepticism—mixed with just the right amount of hope—can protect us from our lust for answers while keeping us open-minded enough to The publishes recognize genuine truth if and when it arrives. over IEEE
References 1. N. Angier, “Dark Days in the Laboratory,” New York Computer Times, book review, 30 June 1996. 2. H. Adams, The Education of Henry Adams, Society Houghton Mifflin, 1918. 3. K. E. Drexler, Engines of Creation: The Coming Era 150 of Nanotechnology, Anchor, 1986. conference 4. D.E. Duncan, “Discover Dialogue: Chemist George Whitesides,” Discover, Dec. 2003, p. 24 proceedings 5. R. Landauer, “The Limits of Scientific Knowledge,” Santa Fe Institute meeting, 25 May 1994. a year. 6. S. Wolfram, A New Kind of Science, Wolfram Media, 2002. 7. F. Hayes-Roth, “The Machine as Partner of the New Professional,” IEEE Spectrum, June 1984, pp. 28-31. for a 8. H.P. Moravec, Robot: Mere Machine to Transcen- dent Mind, Oxford Univ. Press, 1998. preview of 9. R. Kurzweil, The Age of Spiritual Machines: When Computers Exceed Human Intelligence, Oxford Univ. Press, 1998. the latest 10. D.G. Stork, HAL’s Legacy: 2001’s Computer as Dream and Reality, MIT Press, 1997. papers 11. S. Garfinkel, “Happy Birthday, HAL,” Wired, Jan. Not a member? 1997, p. 188.
Join online today! in your field, 12. J.L. Casti, “Lighter than Air,” Nature, 29 Aug. 1996, p. 769. visit 13. G.M. Edelman, Bright Air, Brilliant Fire: On the Mat- ter of the Mind, Basic Books, 1993. 14. E.R. Kandel, “A New Intellectual Framework for Psy- chiatry,” Am. J. Psychiatry, Apr. 1998, pp. 457-469. www.computer.org/proceedings 15. H. Gardner, “Scientific Psychology: Should We Bury
January 2004 43 NotNot AA MemberMember Yet? Yet? Here Is What You’re Missing... Distance Learning Members of the IEEE Computer Society enjoy FREE access to a comprehensive distance learning program for computer professionals. Join today and take any or all of 100 online course titles through our Distance Learning Campus. Subjects include Java, project management, Cisco networks, UNIX, Windows, XML, Oracle, SQL, and more. Magazines and Journals A FREE subscription to Computer magazine is included with your membership. You may also subscribe to other publications in your area of expertise at member discounts. Or subscribe to the IEEE Computer Society Digital Library—21 periodicals and 1200+ conference proceedings—for a complete online resource. Conferences and Workshops Enhance your knowledge and share practical experiences at more than 150 conferences, workshops, and symposia held each year worldwide. Members save at least 25% on registration fees and get advance notice of the meetings. Other Valuable Benefits • FREE membership in your local chapter • FREE e-mail alias of [email protected] • FREE membership in up to four of 40+ Technical Committees • FREE membership in 160+ Standards Working Groups • Member discounts on hundreds of books and conference proceedings Join the IEEE Computer Society Today!
Complete the adjacent membership application today. For fastest service, apply online now at http://computer.org/join 2004 IEEE Computer Society Professional Membership/Subscription Application Membership and periodical subscriptions are annualized to and expire on 31 December 2004. Pay full or half-year rate depending upon the date of receipt by the IEEE Computer Society as indicated below.
Membership Options* FULL YEAR HALF YEAR Payment Information Applications received Applications received All prices are quoted in U.S. dollars 16 Aug 03 - 29 Feb 04 1 Mar 04 - 15 Aug 04 Payment required with application 1 I do not belong to the IEEE, and I want to join just the Computer Society $ 99 ❑ $50 ❑ Membership fee $ ______2 I want to join both the Computer Society and the IEEE: Periodicals total $ ______I reside in the United States $189 ❑ $95 ❑ Applicable sales tax*** $ ______I reside in Canada $170 ❑ $85 ❑ Total $ ______I reside in Africa/Europe/Middle East $166 ❑ $83 ❑ I reside in Latin America $159 ❑ $80 ❑ Enclosed: I reside in Asia/Pacific $160 ❑ $80 ❑ ❑ Check/Money Order**** 3 I already belong to the IEEE, and I want to join the Computer Society. $ 42 ❑ $21 ❑ Charge my: ❑ MasterCard ❑ Visa (IEEE members need only furnish name, address, and IEEE number with payment.) ❑ American Express ❑ Diner’s Club
Are you now or were you ever a member of the IEEE? ______Yes ❑ No ❑ If yes, provide member number if known: ______Card number
______FULL YEAR HALF YEAR Expiration date (month/year)
ISSUES Applications received Applications received ______PER 16 Aug 03 - 29 Feb 04 1 Mar 04 - 15 Aug 04 Signature Add Periodicals** YEAR PRINT ELECTRONIC COMBO PRINT ELECTRONIC COMBO USA-only include 5-digit billing zip code IEEE Computer Society Digital LibraryBEST DEAL n/a n/a $109 ❑ n/a n/a $55 ❑ n/a Computing in Science and Engineering 6 $42 ❑ $34 ❑ $55 ❑ $21 ❑ $17 ❑ $28 ❑ ❑ ❑ ❑ ❑ ❑ ❑ * Member dues include $17 for a 12-month subscription to Computer. IEEE Computer Graphics and Applications 6 $39 $31 $51 $20 $16 $26 ** Periodicals purchased at member prices are for the member’s IEEE Design & Test of Computers 6 $37 ❑ $30 ❑ $48 ❑ $19 ❑ $15 ❑ $24 ❑ personal use only. *** Canadian residents add 15% HST or 7% GST to total. AL, AZ, CO, IEEE Intelligent Systems 6 $37 ❑ $30 ❑ $48 ❑ $19 ❑ $15 ❑ $24 ❑ DC, NM, and WV add sales tax to all periodicals. GA, IN, KY, MD, and MO add sales tax to print and combo periodicals. NY add sales tax to ❑ ❑ ❑ ❑ ❑ ❑ electronic periodicals. European Union residents add VAT tax to IEEE Internet Computing 6 $39 $31 $51 $20 $16 $26 electronic periodicals. IT Professional 6 $40 ❑ $32 ❑ $52 ❑ $20 ❑ $16 ❑ $26 ❑ **** Payable to the IEEE in U.S. dollars drawn on a U.S. bank account. ❑ ❑ ❑ ❑ ❑ ❑ Please include member name and number (if known) on your check. IEEE Micro 6 $37 $30 $48 $19 $15 $24 † Not part of the IEEE Computer Society Digital Library. Electronic access IEEE MultiMedia 4 $35 ❑ $28 ❑ $46 ❑ $18 ❑ $14 ❑ $23 ❑ is through www.ieee.org/ieeexplore. IEEE Pervasive Computing 4 $41 ❑ $33 ❑ $53 ❑ $21 ❑ $17 ❑ $27 ❑ ❑ ❑ ❑ ❑ ❑ ❑ For fastest service, IEEE Security & Privacy 6 $41 $33 $53 $21 $17 $27 apply online at IEEE Software 6 $44 ❑ $35 ❑ $57 ❑ $22 ❑ $18 ❑ $29 ❑ IEEE/ACM Transactions on Computational http://computer.org/join NEW Biology and Bioinformatics 4 $35 ❑ $28 ❑ $46 ❑ $18 ❑ $14 ❑ $23 ❑ NOTE: In order for us to process your IEEE/ACM Transactions on Networking† 6 $44 ❑ $33 ❑ $55 ❑ $22 ❑ $17 ❑ $28 ❑ application, you must complete and IEEE Transactions on: return BOTH sides of this form to the Computers 12 $41 ❑ $33 ❑ $53 ❑ $21 ❑ $17 ❑ $27 ❑ office nearest you: Dependable and Secure Computing NEW 4 $31 ❑ $25 ❑ $40 ❑ $16 ❑ $13 ❑ $20 ❑ Information Technology in Biomedicine† 4 $35 ❑ $25 ❑ $44 ❑ $18 ❑ n/a $22 ❑ Asia/Pacific Office Knowledge and Data Engineering 12 $43 ❑ $34 ❑ $56 ❑ $22 ❑ $17 ❑ $28 ❑ IEEE Computer Society Watanabe Bldg. Mobile Computing 4 $30 ❑ $24 ❑ $39 ❑ $15 ❑ $12 ❑ $20 ❑ † 1-4-2 Minami-Aoyama Multimedia 6 n/a n/a $38 ❑ n/a n/a n/a Minato-ku, Tokyo 107-0062 Japan NanoBioscience† 4 $30 ❑ $24 ❑ $38 ❑ $15 ❑ n/a $19 ❑ Phone: +81 3 3408 3118 Parallel and Distributed Systems 12 $40 ❑ $32 ❑ $52 ❑ $20 ❑ $16 ❑ $26 ❑ Fax: +81 3 3408 3553 Pattern Analysis and Machine Intelligence 12 $44 ❑ $35 ❑ $57 ❑ $22 ❑ $18 ❑ $29 ❑ E-mail: [email protected] Software Engineering 12 $38 ❑ $30 ❑ $49 ❑ $19 ❑ $15 ❑ $25 ❑ Publications Office Visualization and Computer Graphics 6 $34 ❑ $27 ❑ $44 ❑ $17 ❑ $14 ❑ $22 ❑ IEEE Computer Society VLSI Systems† 12 n/a n/a $28 ❑ n/a n/a $14 ❑ 10662 Los Vaqueros Circle IEEE Annals of the History of Computing 4 $31 ❑ $25 ❑ $40 ❑ $16 ❑ $13 ❑ $20 ❑ PO Box 3014 Los Alamitos, CA 90720-1314 USA Phone: +1 714 821 8380 Fax: +1 714 821 4641 Choose PRINT for paper issues delivered via normal postal channels. E-mail: [email protected] Choose ELECTRONIC for 2004 online access to all issues published from 1988 forward. Choose COMBO for both print and electronic. Allow up to 8 weeks to complete application processing. Allow a minimum of 6 to 10 weeks for delivery of print periodicals. IF4A Personal Information BPA Information Enter your name as you want it to appear on correspondence. This information is used by society magazines to verify their As a key identifier in our database, circle your last/surname. annual circulation. Please refer to the audit codes and indicate your selections in the box provided. Male ❑ Female ❑ Date of birth (Day/Month/Year) A. Primary line of business Title First name Middle Last/Surname 1. Computers 2. Computer peripheral equipment 3. Software Home address 4. Office and business machines 5. Test, measurement and instrumentation equipment 6. Communications systems and equipment City State/Province 7. Navigation and guidance systems and equipment 8. Consumer electronics/appliances 9. Industrial equipment, controls and systems Postal code Country 10. ICs and microprocessors 11. Semiconductors, components, sub-assemblies, materials and supplies Home telephone Home facsimile 12. Aircraft, missiles, space and ground support equipment 13. Oceanography and support equipment 14. Medical electronic equipment Preferred e-mail 15. OEM incorporating electronics in their end product (not elsewhere classified) 16. Independent and university research, test and design laboratories and Send mail to: ❑ Home address ❑ Business address consultants (not connected with a manufacturing company) 17. Government agencies and armed forces 18. Companies using and/or incorporating any electronic products in their manufacturing, processing, research, or development activities Educational Information 19. Telecommunications services, telephone (including cellular) 20. Broadcast services (TV, cable, radio) 21. Transportation services (airlines, railroads, etc.) Computer and communications and data processing services First professional degree completed Month/Year degree received 22. 23. Power production, generation, transmission, and distribution 24. Other commercial users of electrical, electronic equipment and services Program major/course of study (not elsewhere classified) 25. Distributor (reseller, wholesaler, retailer) 26. University, college/other education institutions, libraries College/University State/Province Country 27. Retired 28. Others (allied to this field) Highest technical degree received Program/Course of study B. Principal job function 1. General and corporate management 2. Engineering management Month/Year received 3. Project engineering management 4. Research and development management College/University State/Province Country 5. Design engineering management - analog 6. Design engineering management - digital 7. Research and development engineering 8. Design/development engineering - analog Business/Professional Information 9. Design/development engineering - digital 10. Hardware engineering 11. Software design/development 12. Computer science Title/Position 13. Science/physics/mathematics 14. Engineering (not elsewhere classified) Marketing/sales/purchasing Years in current position Years of practice since graduation 15. 16. Consulting 17. Education/teaching Employer name Department/Division 18. Retired 19. Other
Street address City State/Province C. Principal responsibility 1. Engineering or scientific management 2. Management other than engineering Postal code Country 3. Engineering design 4. Engineering 5. Software: science/management/engineering Office phone Office facsimile 6. Education/teaching I hereby make application for Computer Society and/or IEEE membership and 7. Consulting agree to be governed by IEEE’s Constitution, Bylaws, Statements of Policies and 8. Retired 9. Other Procedures, and Code of Ethics. I authorize release of information related to this application to determine my qualifications for membership. D. Title 1. Chairman of the Board/President/CEO 2. Owner/Partner 3. General Manager 4. V.P. Operations Signature Date 5. V.P. Engineering/Director Engineering APPLICATION MUST BE SIGNED 6. Chief Engineer/Chief Scientist 7. Engineering Manager 8. Scientific Manager 9. Member of Technical Staff 10. Design Engineering Manager NOTE: In order for us to process your application, 11. Design Engineer you must complete and return both sides of this form. 12. Hardware Engineer 13. Software Engineer 14. Computer Scientist 15. Dean/Professor/Instructor 16. Consultant 17. Retired 18. Other Professional/Technical COVER FEATURE
2003 Technology Roadmap for Semiconductors
This update to the 2001 ITRS Roadmap shows the industry shifting its focus toward systems on chip, wireless computing, and mobile applications.
Don wo years ago, our report titled “2001 memory—as well as the SoC products they com- Edenfeld Technology Roadmap for Semiconduc- prise. The 2003 ITRS system drivers chapter pre- Intel tors” (Computer, Jan. 2002, pp. 42-53) sents an overarching SoC context for future semi- summarized our assessment of the oppor- conductor products, along with new discussions of Andrew B. T tunities and challenges awaiting the semi- technology requirements for analog/mixed-signal Kahng conductor industry in its continued pursuit of and embedded memory fabrics. University of Moore’s law. The 2003 edition updates this effort California, and reveals a significant shift in the industry’s focus. Systems on chip San Diego Today, the introduction of new technology solu- The SoC product class is a yet-evolving design tions is increasingly application-driven, with prod- style that integrates technology and design elements Mike ucts for different markets using different technology from other system driver classes into a wide range of Rodgers combinations at different times. General-purpose high-complexity, high-value semiconductor prod- Intel digital microprocessors for personal computers ucts. SoC manufacturing and design technologies have been joined by mixed-signal systems for wire- are typically developed originally for high-volume Yervant less communication and embedded applications. custom drivers such as processors, field-program- Zorian Battery-powered mobile devices are as strong a dri- mable gate arrays (FPGAs), or memories. SoCs inte- Virage Logic ver as wall-plugged servers. System-on-chip (SoC) grate building blocks from the other system driver and system-in-package (SiP) designs that incorpo- classes and increasingly subsume the application- rate building blocks from multiple sources are sup- specific integrated circuit (ASIC) category. planting in-house, single-source chip designs. SoCs exist to provide low cost and high integra- tion. Lower implementation cost requires greater SYSTEM DRIVERS reuse of intellectual property as well as platform- The International Technology Roadmap for based design, silicon implementation regularity, or Semiconductors (ITRS) provides quantified, self- other novel circuit and system architecture para- consistent models of canonical products—system digms. Cost considerations also drive the deploy- drivers—that drive the semiconductor industry. ment of low-power process and low-cost packaging These models support extrapolation of future tech- solutions, along with fast-turnaround-time design nology requirements for basic circuit fabrics— methodologies. The latter, in turn, require new stan- processor, analog/mixed-signal, and embedded dards and methodologies for IP description and test,
0018-9162/04/$20.00 © 2004 IEEE Published by the IEEE Computer Society January 2004 47 Logic SRAM Flash E-DRAM CMOS RF FPGA • development of SoC test methodology, includ- FRAM ing test reusability and analog/digital built-in MEMS self-test (BIST). Chemical sensors Electro-optical Electrobiological Multitechnology integration. Considerations such as 98 99 00 01 02 03 04 06 08 10 12 cost, form factor, connection speed and overhead, and reliability drive the need to build heterogeneous systems on a single chip. As process technologists Figure 1. Existing and predicted first integrations of SoC technologies with stan- seek to meld CMOS with MEMS, bio and chemical dard CMOS processes. sensors, and so on, process complexity and produc- tion cost become major constraints. The total cost of processing is difficult to predict for future mate- 1.60 rials and new combinations of processing steps. Dynamic power LOP However, at present, cost considerations limit the 1.40 Dynamic power LSTP number of technologies on a given SoC: Processes Static power LOP are increasingly modular, but the modules are not 1.20 Static power LSTP Memory power LOP generally stackable. Memory power LSTP 1.00 Power for LOP bottom-up CMOS integration with other technologies Power for LSTP bottom-up depends not only on basic technical advances but 0.80 also on being more cost-effective than multiple-die Watts SiP alternatives. Figure 1 shows how first integra- 0.60 tions of each technology within standard CMOS 0.40 processes—not necessarily together with other tech- nologies, and not necessarily in volume produc- 0.20 tion—might evolve. The power challenge. Recent ITRS editions project 0.00 the implications of power management—a grand 2003 2006 2009 2012 2015 2018 challenge for the semiconductor industry—on the achievable space of SoC designs. From the func- Figure 2. Total chip power trend, in watts, for a SoC-LP personal-digital-assistant tionality of a canonical low-power multimedia application, for both low-operating-power (LOP) and low-standby-power (LSTP) PDA application, an architecture of processor, modes. memory, and interface blocks can be inferred. Chip power requirements can then be estimated bottom block interface synthesis, and so on. Heterogeneous up from the implied logic and memory content, and integration requires merging reprogrammable, from the roadmap for process and circuit para- memory, analog and RF, MEMS, and software ele- meters. ments, as well as chip-package-system cooptimiza- Figure 2 shows the bottom-up lower bound for tion. Thus, SoCs provide the driver for convergence total chip power at an operating temperature of of multiple technologies not only in the same sys- 100° C, even given the assumption of multiple tem package, but also potentially in the same man- device technologies integrated within a single core ufacturing process. Overall, SoC designs present to afford greater control of dynamic power, standby many challenges to design, test, process integration, power, and performance. The salient observation and packaging technologies. Most daunting among is that significant advances are required in archi- these challenges are tecture and design technology to meet the system targets of 100 mW peak and 2 mW standby power. • design productivity improvement of greater than 100 percent per node, including platform- Mixed-signal chips based design and integration of programma- Analog/mixed-signal (AMS) chips at least par- ble logic fabrics; tially deal with input signals whose precise values • power management, especially for low-power, matter. This broad class includes RF, analog, ana- wireless, multimedia applications; log-to-digital, and digital-to-analog converters— • system-level integration of heterogeneous tech- and, more recently—many mixed-signal chips in nologies, including MEMS and optoelectron- which at least part of the chip design must measure ics; and signals with high precision. These chips have design
48 Computer 22 Super 1 kW audio 20 1 W 18 1 mW and process technology demands that differ from those for digital circuits. Technology scaling is not 16 Audio GSM necessarily helpful for analog circuits because deal- 14 ADS_ basestation ing with precision requirements or signals from a GSM fixed voltage range is more difficult with scaled 12 1 µW Cable voltage supplies. In general, AMS circuits and Resolution (bit) 10 W-CDMA DTV process technologies present severe challenges to Blue- VDSL tooth WLAN 8 Tele- Video Storage cost-effective CMOS integration. phony The need for precision also affects tool require- 6 ments for analog design. Digital circuit design cre- Inter- 4 connectivity ates a set of rules for correct logic gate function; as long as developers follow these rules, precise cal- 1 kHz 10 kHz 100 kHz 1 MHz 10 MHz 100 MHz 1 GHz culation of exact signal values is unnecessary. Signal bandwidth Analog designers, on the other hand, must concern themselves with many second-order effects to obtain the required precision they require. Relevant Figure 3. Recent analog-to-digital converter (ADC) performance needs for issues include coupling and asymmetries. Analysis important product classes. tools for these issues are mostly in place but require expert users, while synthesis tools are immature. occupy less area at the expense of increased Manufacturing test for AMS circuits poses a prob- total cost; lem that remains essentially unsolved. • technology choices can impact design cost by Most analog and RF circuitry in today’s high- introducing greater risk of multiple hardware volume applications resides in SoCs. The economic passes; regime of a mainstream product is usually highly • parametric yield sensitivities can impact man- competitive: It has a high production volume and ufacturing cost; and hence requires a high level of R&D investment; • a SiP multidie solution can be more cost- thus, technology requirements drive mixed-signal effective than a single SoC solution. technology as a whole. Mobile communication platforms are the highest-volume circuits driving Such considerations make mixed-signal design the needs of mixed-signal circuits. cost estimation difficult. We can attempt to quan- Mixed-signal evolution. The interplay between cost tify mixed-signal cost by first restricting our atten- and performance determines mixed-signal driver tion to high-performance applications, since these evolution, including its scope of application. also drive technology demands. Because analog fea- Although mixed-signal performance is an impor- tures are embodied as high-performance passives tant metric, cost of production is also critical for or analog transistors, that area can be taken as a practical deployment of AMS circuits. Together, proxy for cost. Since improving digital density cost and performance determine the sufficiency of drives transistor scaling, analog transistors can given technology trends relative to existing appli- simply follow, and it isn’t necessary to specifically cations. They also determine the potential of given address their layout density. At the same time, technologies to enable and address entirely new embedded passives determine total area in most applications. current AMS designs, whose area consumption Cost estimation. In mixed-signal designs, chip area dominates a system’s mixed-signal costs. determines only one of several cost factors. The Estimating technology sufficiency. Figure 3 shows area that analog circuits occupy in a SoC typically requirements for analog-digital converter (ADC) cir- ranges from 5 to 30 percent. Economic forces to cuits in terms of a power-performance relationship. reduce mixed-signal area are therefore not as strong Under conditions of constant performance, a as those affecting logic or memory. Related con- straight line with slope of −1 represents constant siderations include the following: power consumption. Increasing performance— achievable with better technology or circuit design— • sometimes, shifting the system partitioning is equivalent to a shift of the power consumption between the analog and digital parts can lines toward the upper right. The data shows a very reduce the analog area; slowly moving technological barrier line for ADCs • introducing high-performance analog devices for a power consumption of 1 W. Most of today’s increases process complexity so that solutions ADC technologies lie below the 1 W barrier, and
January 2004 49 near-term solutions for moving the barrier Embedded memory more rapidly are unknown. SoC designs contain an increasing number and The challenge Although ADC performance has improved variety of embedded RAM, ROM, and register file for compound at a rate adequate for handset applications, it memories. Interconnect and I/O bandwidths, semiconductors is clearly has failed to do so for applications design productivity, and system power limits all to increase the such as digital linearization of GSM base sta- point to a continuing trend of high levels of mem- number of devices tions, or handheld and mobile high-data-rate ory integration in microelectronic systems. digital-video applications. Assuming progress Driving applications for embedded memory tech- per unit area and at recent rates, manufacturing ADCs with nology include code storage in reconfigurable integrate them with adequate performance in volume will not applications, data storage in smart or memory CMOS processing. occur until 2010 or later. Silicon and silicon cards, and the high memory content and high- germanium technologies have the necessary performance logic found in gaming or mass stor- bit resolution, but not the speed, to provide age systems. such performance now; on the other hand, Opportunities and constraints. The balance between III-V compound semiconductor technologies have logic and memory content reflects overall system the speed but not the bit resolution. The challenge cost, power and I/O constraints, hardware-soft- for compound semiconductors then is to increase ware organization, and overall system and mem- the number of devices per unit area and to integrate ory hierarchy. With respect to cost, the device them with CMOS processing. performance and added mask levels of monolithic Enabling new applications. In the semiconductor logic-memory integration must be balanced against industry, improving technology and design perfor- chip-laminate-chip or other SiP integration alter- mance enables new applications, which pushes the natives. Levels of logic-memory integration also industry into new markets. Researchers also can reflect tradeoffs in hardware-software partitioning use mixed-signal design analysis to estimate design as well as code-data balance. needs and feasibility for future applications and I/O pin count and signaling speeds determine new markets. how system organization trades off bandwidth ver- Alternatively, when they know the specifications sus storage: Memory access can be made faster at of a new product, researchers can design the tech- the cost of peripheral overhead by organizing mem- nology needed to fulfill these specifications and ory in higher or lower bank groups. Access speed estimate the timeframe in which the semiconduc- also depends on how pin count and circuit com- tor industry will be able to build that product with plexity are balanced between high-speed, low-pin- acceptable cost and performance. The ability to count connections and higher-pin-count, lower- build high-performance mixed-signal circuitry at speed connections. low cost will continuously drive the semiconduc- In the traditional processor architecture domain, tor industry to develop new products and mar- memory hierarchy is crucial in matching processor kets. speed requirements to memory access capabilities. Mixed-signal challenges. For most of today’s This has led to the introduction of several layers of mixed-signal designs, a voltage difference repre- hardware-controlled caches between main mem- sents the processed signal so that the supply ory and foreground memory in the processor core. voltage determines the maximum signal. Typically, one physical cache memory is present at Decreasing supplies—a consequence of constant- each layer. However, the choice of hierarchy also field scaling—mean decreasing the maximum has strong implications for power. Conventional achievable signal level. This strongly impacts architectures increase performance largely at the mixed-signal product development for SoC solu- cost of energy-inefficient control overheads by tions. Typical development time for new mixed- using, for example, prediction and history mecha- signal parts takes much longer than for digital and nisms and extra buffers arrayed around highly asso- memory parts, which makes lack of design ciative caches. From the system viewpoint, the resources another key challenge. embedded multimedia and communication appli- Overall, the most daunting mixed-signal chal- cations that dominate on portable devices can profit lenges are decreasing supply voltage, increasing rel- more from software-controlled and -distributed ative parametric variations, increasing numbers of memory hierarchies. analog transistors per chip, increasing processing Different layers of the hierarchy require differ- speed, increasing crosstalk, and shortage of design ent access modes and internal partitionings. The skills and productivity. use of page, burst, and interleaving modes, and the
50 Computer physical partitioning into banks, subarrays, and counts that are enabled by smaller feature divided-words and bitlines generally must be opti- sizes and spurred by consumer demand for Embedded software, mized per layer. Increasingly dominant, leakage increased functionality, lower cost, and power constraints also lead to more heterogeneous shorter time to market. Together, these chal- manufacturing test, memory hierarchies. lenges imply exponentially increasing design and design Embedded memory challenges. At the circuit level, process complexity. New paradigms are verification continue scaling issues such as amplifier sense margins for needed in every area of design technology— to escalate into SRAM and decreased I drive currents for DRAM design process, system-level design, logical- on crises, even as present two clear challenges. Smaller feature sizes physical-circuit design, design verification, allow integration of more devices into a single and design for testability—as we move to traditional product, which leads to increased parametric yield the 65-nm technology node and beyond. complexity loss with respect to both noise margins and leak- Examples from the 2003 ITRS include: challenges persist. age power. Future circuit topologies and design methodologies will need to address these issues. • Communications-centric design. As it Another challenge, error tolerance, becomes becomes impossible to move signals severe with process scaling and aggressive layout across a large die within one clock cycle, or densities. Embedded memory’s soft-error rate to run control and dataflow processes at increases with diminishing feature sizes and affects the same clock rate, design will likely shift both embedded SRAM and DRAM. Moving bits to asynchronous or globally asynchronous and in nonvolatile memory can also suffer upsets. locally synchronous styles. A new focus on Particularly for highly reliable applications such as on-chip communication architectures and pro- those in the automotive sector, error correction will tocols will usher in an era of communications- remain a requirement that entails tradeoffs of yield centric design. and reliability against access time, power, and • Robustness. Network-oriented paradigms, as process integration. Finally, cost-effective manu- well as yield and synchronization issues, sug- facturing test and BIST is a critical SoC requirement gest that future SoC design will contend with for both large and heterogeneous memory arrays. potentially lossy communication and node failures, placing a new focus on robustness. DESIGN For example, implementing a completely spec- Previous ITRS editions documented a design pro- ified function in an inherently imperfect fab- ductivity gap in which the number of available tran- ric demands new approaches to design, sistors grows faster than the ability to meaningfully mapping, and verification. design them. Yet, investment in process technology • Scalability. Tools that operate at the lowest has far outstripped investment in design technol- abstraction levels—such as logical and physi- ogy. Fortunately, progress in design technology con- cal design and verification tools—face instance tinues: The 2003 ITRS estimates that the design cost complexities that at least double with each of the low-power SoC PDA is approximately $20 technology node. Scalability requires new million in 2003, versus $630 million had design ways to manage data, search solution spaces, technology improvements since 1993 not been real- and map optimizations onto distributed and ized. However, embedded software, manufacturing parallel computational resources. Construct- test, and design verification continue to escalate into by-correction methodologies and the rise of crises, even as traditional complexity challenges per- reuse demand tool runtimes that are propor- sist. In the 2003 ITRS, the design chapter includes tional not to the size of their inputs, but rather new material on AMS-specific design technology, to the size of changes from the previous ver- as well as the future evolution of design processes. sion of the inputs. • Cost optimization. Multitechnology integra- Facets of the complexity challenge tion in SoC or SiP brings analog and mixed- As our January 2002 ITRS article described signal into the mainstream, as well as codesign (Computer, pp. 48-49), design technologists are of die, package, and substrate levels. System caught between two basic challenges. Silicon com- cost optimization must span many degrees of plexity refers to the emergence of nanometer-scale freedom: multiple-die and stacked-die options, variability and physical effects that place long-stand- package and board interconnects, IP reuse ing paradigms at risk, while system complexity across process generations, and the use of refers to the exponentially increasing transistor reprogrammable blocks.
January 2004 51 • Cooptimization. Implementation tleneck and leveraging reusable, retargetable ana- tools create logic and timing struc- log IP generators. More general techniques can aug- AMS scaling and tures concurrently with constraint ment today’s specialized automatic circuit syntheses migration faces budgets and spatial embedding. for particular circuit classes. Syntheses must also many challenges, For example, today’s register trans- handle future regimes of increased manufacturing including decreasing fer level (RTL) floorplanning and variability using, for example, hybrid analog- supply voltages and global interconnect planning tools digital compensation for device mismatch. define repeater insertion and pipe- In design verification, AMS circuits require increasing relative lining, along with detailed layout checking to specification, not structure. New veri- parametric of global signals. As logical-physi- fication solutions must include statistical tech- variations. cal and layout-clock-test synthesis niques, better compact models that speed up unifications continue, a near-term simulation while increasing accuracy, and new goal is cooptimization of timing acceptance criteria. AMS designs also force the structure, signaling strategies, logic issue of performing hybrid-systems verification into optimization, placement, and routing in a the near term. This creates an immediate challenge single environment. to improve the current ad hoc approaches and find a path toward more powerful techniques. Analog/mixed-signal challenges Analog and RF circuits differ from digital circuits Design process trends in that they lack quantized information represented Four major trends govern future leading-edge in a defined range of voltage levels for the two states chip design processes and their supporting design of a bit and during a defined discrete time interval. system structures: Rather, analog and RF circuits process continuous signals to a much higher degree of precision or • Tight coupling. Design processes are evolving smaller tolerance both in time and amplitude. into collections of modular applications that Therefore, factors such as linearity, noise, parasitic operate concurrently and share design data in elements, and electrical nonuniformity directly memory. In modern methodologies, optimiza- cause distortion and noise in analog or RF signals. tion loops can no longer contain slow file Speed issues, or simply that a signal-recovery accesses, and the plethora of design issues circuit produces more noise and distortion than it requires simultaneous optimization of multi- prevents, make these factors much more challeng- ple criteria. Further advances will be needed ing and less straightforward in the analog domain. to avoid noise problems, minimize power dis- AMS scaling and migration faces many challenges, sipation, and ensure manufacturability. including decreasing supply voltages; increasing rel- • Design for manufacture. More intelligent data ative parametric variations; increasing numbers of preparation requires new characterizations of analog transistors per chip; increasing signal, clock, manufacturing process and cost tradeoffs. and intrinsic device speeds; increasing leakage and While inspection and repair form the largest crosstalk in SoC integration; and a shortage of component of mask-making cost and delay, design skills and automation. manufacturers perform the process without In system-level design, the critical AMS chal- insight into design intent, wasting effort in sat- lenges are analog circuit nonscalability and analog isfying identical tolerances for every shape. We behavioral modeling and synthesis. Automated need a standard framework to communicate analog circuit synthesis and optimization is needed, design data criticality to the mask-making along with language-level modeling methodologies process and to feed manufacturing complex- that permit analysis of overall system function and ity back to the design process. interfaces with other integrated technologies. Issues • Increasing abstraction levels. Most design include coping with vastly varying time scales in today takes place at the gate level for greater simulation, creating heterogeneous test benches and productivity and at the RTL to specify design ensuring coverage, achieving systematic top-down in a modern flow. Continued designer pro- constraint propagation, and mixing functional and ductivity improvements require an emerging structural representations. system-level of design, well above RTL. Higher Analog synthesis provides the key challenge in abstraction levels allow verification to discover logical, physical, and circuit design. Scalable SoC problems much earlier in the design process, design requires eliminating the analog design bot- which reduces time to market and lowers costs.
52 Computer • Increasing automation levels. Using new mod- In general, solutions to silicon complexity els to specify design intent leads to opportu- challenges will entail restrictions on design In general, solutions nities for introducing other tools, such as rules as well as continued improvement in synthesis. This trend replaces designer guid- analyses. Potential solutions to system com- to silicon complexity ance with constraint-driven optimization to plexity challenges will strive for increased challenges will reduce the number of iterations in later capacity, use of hierarchical methods, and a entail restrictions process steps. In future technology nodes, the higher abstraction level of design. Joining on design rules system-level specification must include both these two trends is the evolution of the design software and hardware and become the con- process and design system architecture itself, and continued trolling representation for constraint-driven as depicted in Figure 4. Tool integrations must improvement in implementation. minimize unproductive data translation time analyses. and data redundancy and support synthesis- Detailed implications of these precepts are given analysis tool architectures that execute con- in the context of a new ITRS canonical design flow, currently on distributed or SMP platforms. which serves as a framework for specification of Some specific predictions for the future follow. future design technology innovations. For exam- Designer productivity will improve through mul- ple, in system-level design the simplification of tilevel automation, increased reuse, and freedom mixed hardware and software system specification, from choice. To systematically advance design tech- verification, and implementation requires a new nology in this direction, design must be treated as level of abstraction above the familiar RTL. This science, not art, with designers adopting a measure- necessitates the following advances: to-improve mindset. Eventually, predictive models of application- or driver-specific silicon implemen- • Reuse-based design. Reusable, high-level func- tation will be needed, while complementary inno- tional IP blocks offer the potential for pro- vations should measure and improve design ductivity gains estimated to be at least 200 technology productivity. percent. Preverification and reusable tests Approaches to reduce time to market will include reduce design complexity, and libraries of use of reprogrammable and structured-ASIC fab- reusable software modules can speed embed- rics, reuse of predesigned cores and platform archi- ded software development. tectures, and pervasive automation. New tools will • Platform-based design. An extension of core- be required to support a find-and-try style of reuse- based design creates highly reusable groups of centric design-space exploration and design opti- cores to form a complete hardware platform, mization. Development and formalization of design further simplifying the SoC design process. rules that, if followed, assure reusability, will • System-level verification. Raising the abstrac- require associated design and analysis software. tion level will require a single notation for sys- In the long term, building on the design for man- tem-level design. Several years of experimenta- ufacturability trend, we will need restricted design tion with C, C++, and Java variants has led to rules, as well as the integrated design and manu- the recent emergence of SystemC as an option facturability optimizations indicated in Figure 4. for building interoperable system models of Even today, the focus on yield-driven layout makes hardware and software for simulation. manufacturability a standard design criterion. • Microarchitecture synthesis. Although system synthesis is extremely difficult, progress will TEST likely start with automatic creation of an effec- The dramatic increase in SoC and SiP designs, tive RTL specification from a slightly higher- largely targeted to commodity consumer applica- level representation of the hardware in tions, has consequently increased pressure to microarchitecture-specification form. reduce the cost of test for mixed-technology • Hardware-software cosynthesis. Achieving the designs. These designs break the traditional barri- best overall solution will require the ability to ers between digital, analog, RF, and mixed-signal concurrently synthesize a hardware and soft- test equipment capability requirements, resulting ware implementation. By 2007, we anticipate in a trend toward highly configurable, one-plat- that an automated step will replace the manual form-fits-all test solutions. The first generation of process of mapping a behavioral specification this equipment combined leading-edge technology to a software program and a hardware from all segments, with the consequence of increas- microarchitecture. ing test cost due to the high capital cost of this
January 2004 53 Past (180 nm) Present (130–90 nm) Future (65 nm–)
System System System System System System design design model design model model
Specification Functional Specification performance Performance Functional HW/SW HW/SW testability model verification optimization optimization verification
Functional SW Cockpit verification SW Software optimization Autopilot RTL Software RTL Optimize Analyze optimization Performance HW/SW Communication Performance testability SW Timing HW/SW Synthesis verification Logic Power Circuit data Noise + Timing analysis model Place Test + Placement optimization Cockpit EQ check Wire Mfg. Autopilot Other Repository Other Performance Optimize Analyze File testability verification Logic Communication Timing Masks EQ check Place Power Wire Data Noise • Multiple design files converge into one efficient data model Place/wire EQ check Other model Test • Disk accesses are eliminated in critical + Timing analysis Other + Logic optimization methodology loops Repository • Verification of function, performance, testability, and other design criteria all move to earlier, higher levels of abstraction followed by File Masks • Equivalence checking • Assertion-driven design optimizations Masks • Industry standard interfaces for data access and control • Incremental modular tools for optimization and analysis
Figure 4. Required approach. Low-cost-equipment solutions targeting approach offers several potential advantages. It evolution of design DFT-enabled devices do not scale into the mixed- would system technology space today. The next logical step is to architecture. increase test system configurability and flexibility • focus research and development investment, to achieve a more appropriate cost performance both dollars and effort, on the test instrument point—which is leading to a fundamental shift in itself rather than test infrastructure; test equipment architecture. • base differentiation on test capability rather This revision of the Roadmap finds the test- than platform; equipment industry at the beginning of a signifi- • focus supplier efforts on developing solutions cant shift from traditional test-architectures to within their particular core competency, reduc- universal-slot architectures with high levels of test- ing cost and speeding time to market; instrument encapsulation and modularity. The con- • reduce the investment in reengineering infra- tinued evolution of technology and increased structural elements; and integration level of design components has enabled • eliminate the need for each supplier to be this shift. In many cases, a single FPGA or ASIC everything to everyone. can take the place of entire electronics subsystems. Encapsulation and modularity have led to the Ultimately, the industry will drive this approach’s open-architecture concept: the ability to mix and success. No one can dispute, however, that the next match test instruments from multiple suppliers into several years represent a turning point for many a single tester hardware and software environment. test-solution suppliers. Regardless of the open- This concept presents significant business model architecture approach’s success, emerging plat- challenges to the test equipment industry, which forms offer a new level of capability, flexibility, and today is based on full vertical integration and pro- longevity that will significantly impact the indus- prietary platforms. However, an open-architecture try and test deployments. Quite possibly, the next
54 Computer significant equipment selection decision could iden- penetration of high-speed serial protocols tify the test platform deployed for all products into the microprocessor, ASIC, and SoC mar- through the next decade. kets in the form of multilane buses has accel- The high complexity While for many years the cost of the tester has erated dramatically. This trend brings a of SoC design overwhelmed all other parameters—with the single complex test problem, previously limited to creates design-and- exception of throughput—significant progress has the high-speed networking environment, into test development- been made to address this aspect of the overall the mainstream. productivity and device-test cost equation. Focus is shifting to address Key lessons from this market segment indi- the numerous secondary costs that now become cate the need to execute extensive testing such test-quality large contributors. This analysis must encompass as jitter tolerance and jitter transfer on these challenges. all aspects of test cost, including the design non- interfaces. Such testing is done today in the recurrent engineering associated with design for test, analog domain through a rack-and-stack or handler or prober equipment, device interface hard- mixed-signal tester approach. These solutions ware, and facilities cost, among many other factors. impose significant manufacturing cost considera- tions due to test time and equipment capital cost, SoC test and they support a relatively limited number of SoC test is dependent on a highly structured DFT high-speed serial ports on a single device. As these methodology to enable observability and control- interfaces proliferate to many ports on a single lability of individual cores. Increasingly, SoC design device, the traditional analog test approach will fail will rely on a database of preexisting IP cores that due to the scalability of analog instrumentation. encapsulate the design itself, interfaces to other As the frequency of these interfaces and the num- blocks, and test. ber of interfaces on a single design continue to A fundamental challenge of SoC test is the need increase, alternative equipment solutions and test to combine test requirements from multiple sources methods will need to be developed to enable cost- with differing testability approaches and methods. effective, high-coverage engineering and manufac- Opportunities exist to define standards for test to turing test. conform to a hierarchical methodology; these stan- dards are most easily imposed on internally Reliability screens designed cores. However, when IP is purchased or The test process is responsible for the screening licensed from a third party, it is typically the test of manufacturing defects that affect device func- methodology that must adapt. Many electronic tionality, performance, and reliability to reduce cus- design automation (EDA) tools already leverage a tomer perceived defects per million. A portion of standard format for logic designs. This standard the test flow is dedicated to the acceleration of must be extended to other core types, such as ana- latent defects that do not appear as test failures but log circuits. would manifest as longer-term reliability failures. Structured use of IP core wrappers and test access Traditional techniques for screening of reliability mechanisms must be developed for testing of indi- failures include IDDQ, burn-in, and voltage stress. vidual cores within a SoC. These should be devel- The effectiveness of these methods is challenged oped carefully to enable functional, at-speed by the continued device scaling with each successive parametric and interconnect testing. Further, these process generation. Increases in device quiescent methods should be standardized with the interface current in the off state is raising the level of back- language for interoperability of EDA tools. One ground current to the milliampere and, in some such effort is the Core Test Language development cases, ampere range. These levels of background within the IEEE P1500 standard. The high com- current increase the difficulty of identifying plexity of SoC design creates design-and-test devel- microampere to milliampere level IDDQ fault cur- opment productivity and test-quality challenges. rents. Continued extension of techniques such as
EDA tools and data interchange standards must be delta-IDDQ that have enabled extensions into cur- developed to aid management of this complexity. rent process generations is uncertain. At the same time, the effectiveness of voltage and High-frequency I/O temperature acceleration methodologies used by High-speed serial interfaces have been used in burn-in and voltage stress is declining due to the the communications market segment for many reduced margin between operational and overstress years. While the communications market is conditions. Costs associated with burn-in tech- expected to maintain a significant frequency lead, niques continue to rise and, in some cases, now
January 2004 55 dominate manufacturing costs for high-power that the resulting effort is indeed reduced over exist- products. ing methods and that pre- and postprocessing of data The increasing cost and declining effectiveness do not eliminate throughput gains. of current techniques for latent defect acceleration combine to create one of the most critical chal- lenges facing the industry for future process gener- revious ITRS editions documented a design ations. Extensions to current techniques may prove productivity gap, with the number of avail- adequate for the next several years, but funda- P able transistors growing faster than the abil- mental research in the development of new method- ity to meaningfully design them. Yet, investment in ologies is required. process technology has by far dominated invest- ment in design technology. Automated test program generation The good news is that enabling progress in design Correct-by-construction test programs and pat- technology continues. The bad news is that soft- terns have long eluded test development teams. ware can account for 80 percent of an embedded Increasing design complexity and demands on team system’s development cost, test cost has increased productivity require significant improvement in test exponentially relative to manufacturing cost, and program generation automation to limit or reduce verification engineers outnumber design engineers time-to-market impact. The EDA industry has on microprocessor project teams. Overcoming developed and deployed many tools intended to aid many existing design technology gaps will require the entire process of test development, content cre- a concerted effort by the entire industry. ation, and equipment translation. However, device makers seldom have a homogeneous environment provided by a single EDA supplier, and the general Acknowledgments lack of interoperability standards among tools cre- We acknowledge the efforts of the many indi- ates significant challenges requiring effort by the viduals who contributed to making the 2003 edi- device maker to enable automation. Cooperation tion of The International Technology Roadmap for among EDA and test equipment suppliers is increas- Semiconductors a successful endeavor. ing, and a focus on tool interface and interoper- ability standards is growing. The challenges increase Don Edenfeld is responsible for test equipment def- with every process generation and related growth inition and selection at Intel. He received an MS in in design integration. electrical engineering from the University of Virginia. Achieving full automation of test-program-gen- Edenfeld is a member of the IEEE Computer Soci- eration will require increased standardization of the ety. Contact him at [email protected]. test-equipment software environment itself. Histor- ically, each equipment supplier has taken a holistic Andrew B. Kahng is a professor of computer sci- and proprietary approach to definition and devel- ence engineering and electrical and computer engi- opment of its specific software environment. Similar neering at the University of California, San Diego. to EDA, most device makers do not have a homo- He received a PhD in computer science from the geneous test environment of equipment provided by University of California, San Diego. He is a mem- a single supplier. Increased standardization of the ber of the ACM, the IEEE, and SPIE. Contact him test-equipment software environment where appro- at [email protected]. priate would lower the entry barrier for suppliers as well as simplify the porting of test content between Mike Rodgers is responsible for test technology platforms as required by the device maker. integration at Intel. He received a BS in electrical Today’s environment of platform-unique supplier engineering from the University of Illinois. Rodgers software solutions and homegrown tools for equip- is a member of the IEEE Computer Society. Con- ment programming, automation, and customization tact him at [email protected]. will drive unacceptable growth in test- development engineering and factory-integration efforts. Auto- Yervant Zorian is the vice president and chief mation of common tasks and decreasing test-plat- scientist of Virage Logic. He received a PhD in form integration time demand a focus on standards electrical engineering from McGill University. to enable more efficient use of resources in line with Zorian is a Fellow of the IEEE and a member shrinking product development life cycles. New tool of the IEEE Computer Society. Contact him at development must comprehend the end use to ensure [email protected].
56 Computer IEEE COMPUTER SOCIETY LIBRARY SUBSCRIPTION PLAN- ELECTRONIC
The Premier Collection of Online Computer Science Publications
IT & SECURITY 22 peer-reviewed computer science and engineering periodicals ARTIFICIAL INTELLIGENCE More than 100,000 computing articles and papers